Luke Kehoe leads Ookla’s research and thought leadership efforts in Europe.
An electronic engineering alumnus of University College Dublin, Luke has extensive experience collaborating with mobile operators, telecoms vendors, and government agencies in research and advisory roles across Europe. He has contributed to internationally recognised thought leadership publications in areas such as 5G, IoT, open RAN, and edge computing, working with prestigious organisations like the Telecom Infra Project and the World Economic Forum.
New wireless silicon in the iPhone 17 family delivers material performance improvements over predecessors, pushing it ahead of many Android flagship devices in Wi-Fi.
If the last few smartphone releases were defined by cellular milestones, 2025 has quietly become the year of Wi‑Fi. Apple’s first custom networking chip, the N1, arrives in the iPhone 17 family, while Android flagships (meaning companies’ top-of-the-line models) have leaned into Wi-Fi 7 and 6 GHz with enhanced capabilities made possible by 320 MHz channels. The primacy of Wi-Fi performance in the everyday user experience and the proliferation of new form factors mean device manufacturers are competing more intensely for access to the best networking silicon.
Using global, crowdsourced Speedtest Intelligence® data from the six weeks after the iPhone 17 family of devices hit stores, we compared the performance of Apple’s N1 with its Broadcom-based predecessor and leading Android flagships using Wi-Fi silicon from Qualcomm, MediaTek and Broadcom.
Key Takeaways:
Apple’s N1 chipset is a substantial upgrade. The iPhone 17 family delivers a clear step-change in Wi-Fi performance vs. the Broadcom-based iPhone 16 lineup, with faster download and upload speeds across every region. Globally, median download and upload speeds on the N1 were each up to 40% higher than on its predecessor.
Google’s Pixel 10 Pro and iPhone 17 families jostle for Wi-Fi leadership. The Pixel 10 Pro recorded the highest global median download speed at 335.33 Mbps during the study period, marginally edging out the iPhone 17 family at 329.56 Mbps. The pattern flips at the 10th percentile (worst-case), where the iPhone 17 family leads globally with 56.08 Mbps, just ahead of the Pixel 10 Pro family at 53.25 Mbps.
Xiaomi’s 15T Pro delivers the strongest upload and latency performance. Based on MediaTek Wi-Fi silicon integrated with the Dimensity 9400(+) platform, the 15T Pro performed strongest in 90th-percentile (best-case) download speed at 887.25 Mbps, upload speed at the 10th, median and 90th percentile levels and median multi-server latency (15 ms) globally.
Huawei’s Pura 80 family suffers from lack of 6 GHz support but remains competitive on non-6 GHz networks. Based on a “self-developed chip-level collaboration” (likely from HiSilicon), it lags other flagships in download and upload speeds, with a particularly acute gap at the 90th percentile where the absence of 6 GHz support hurts peak performance. Notwithstanding this, when looking only at non-6 GHz samples, the Pura 80 family is more competitive and, on Wi-Fi 6, delivers the second-fastest upload speeds at the 90th percentile (603.61 Mbps) in Southeast Asia against Android flagships.
Wi-Fi 7 and 6 GHz are force multipliers for flagship Wi-Fi silicon, though adoption remains regionally skewed. Across Android families, median 6 GHz download speeds were at least 77% faster than 5 GHz, and the step from Wi-Fi 6 to Wi-Fi 7 delivered a similar lift. In North America, flagship Android users spend much more time on 6 GHz networks, with the Galaxy S25 family showing over 20% of Speedtest samples on 6 GHz, compared with about 5% in Europe and Northeast Asia and just 1.7% in the Gulf region.
Methodological note: This analysis uses Speedtest® data collected from September 19 to October 29, 2025. The included Wi-Fi 7-capable devices are listed below. For each device family, the results represent the aggregate of all devices in that family:
Apple iPhone 16 family (iPhone 16, 16 Plus, 16 Pro, 16 Pro Max)
Apple iPhone 17 family (iPhone Air, iPhone 17, iPhone 17 Pro, iPhone 17 Pro Max)
Samsung Galaxy S25 family (Galaxy S25, S25+, S25 Ultra)
Google Pixel 10 Pro family (Pixel 10 Pro, Pixel 10 Pro XL)
Huawei Pura 80 family (Pura 80 Pro, Pura 80 Ultra)
Xiaomi 15T Pro
Vivo X200 Pro
Oppo Find X8 Pro
Apple’s N1 focuses on tighter hardware-software integration rather than chasing peak capability
The arrival of the N1 marks the next ambitious step in Apple’s multi-year plan to bring the last major piece of the iPhone’s wireless stack in-house. By moving off Broadcom-supplied parts, Apple gains tighter control over mission-critical silicon, reduces supplier dependence and pricing exposure and creates a reusable radio platform that can scale across iPhone, Mac, iPad, Watch and Home devices.
Technically, the N1 is a single-die chip that integrates Wi-Fi 7, Bluetooth 6 and Thread radios. Aside from the step up from Bluetooth 5.3 to 6 and Apple’s claim that tighter hardware-software integration improves features like AirDrop and Personal Hotspot, the N1’s Wi-Fi capabilities appear, on paper, virtually identical to its Broadcom-based predecessor.
This continuity in Wi-Fi specifications is notable because it means the N1 is capped at 160 MHz channels and lacks support for 320 MHz operation and thus the peak link rates (or PHY speeds) available with flagship silicon from vendors such as Qualcomm and MediaTek.
In practical terms, this should limit the N1’s peak performance in markets that allow the full 6 GHz band, like the US, which offers up to three non-overlapping 320 MHz channels. It should also limit performance (although potentially to a lesser degree) in regions that allow only the lower 6 GHz block, like the EU and UK, which offer just one non-overlapping 320 MHz channel.
iPhone 17 family delivers a clear step up in Wi-Fi performance over its predecessors
Analysis of Speedtest Intelligence data shows that, despite the similar headline specifications between the Broadcom-based iPhone 16 family and the N1-powered iPhone 17, the 17 delivers a clear step-change in real-world Wi-Fi performance. New devices often appear to outperform in their early weeks, partly because early adopters skew toward wealthier markets with more capable Wi-Fi networks. However, the consistency and magnitude of the iPhone 17’s lead indicate this is not a launch-period skew but a genuine improvement.
iPhone 17 Family Delivers Step-Change in Wi-Fi Performance Globally
Speedtest Intelligence® | Sept 19 – Oct 29, 2025. Regional.
To ensure the gains are not a simple country-mix artifact, we matched markets where both families exhibited the most samples during the study period. Across all of those countries analysed, including major markets such as the US, UK, Germany, Japan, Italy and India, the iPhone 17 outperformed the iPhone 16 on download performance. This pattern holds across markets with very high absolute speeds (e.g., France) and more typical markets alike, pointing to genuine device-side improvements.
N1 Silicon is Driving Wi-Fi Gains Across Major Markets
Speedtest Intelligence® | Sept 19 – Oct 29, 2025. Country-level.
The iPhone 17 family delivered higher download and upload speeds on Wi-Fi compared to the iPhone 16 across every studied percentile (10th, median and 90th) and virtually every region. During the study period, the iPhone 17 family’s global median download of 329.56 Mbps was as much as 40% higher than the iPhone 16 family’s 236.46 Mbps. Upload speeds improved similarly, jumping from 73.68 Mbps to 103.26 Mbps.
iPhone 17 Family Sees Biggest Upload Gains in Asia
Speedtest Intelligence® | Sept 19 – Oct 29, 2025. Regional
Notably, the N1 delivers a far bigger generational uplift at the 10th percentile than at the 90th, implying Apple’s custom silicon lifts the floor more than the ceiling, a pattern we also saw in our analysis of the in-house C1 modem’s cellular performance.
iPhone 17 Family is Stronger in Tough Wi-Fi Conditions
Speedtest Intelligence® | Sept 19 – Oct 29, 2025. Regional.
This means the N1 appears to deliver a more consistent experience across a wider range of environments, in particular uplifting performance under challenging Wi-Fi conditions. Specifically, 10th-percentile speeds on iPhone 17 were over 60% higher, versus just over 20% at the 90th percentile.
Singapore and France Lead Global iPhone 17 Speeds
Speedtest Intelligence® | Sept 19 – Oct 29, 2025. Country-level. iPhone 17 family.
At a regional level, iPhone 17 users enjoyed the highest median download speeds in North America at 416.14 Mbps (up from 323.69 Mbps on the iPhone 16 family), mainly due to greater 6 GHz use. At a country level, meanwhile, iPhone 17 users in Singapore (613.80 Mbps) and France (601.46 Mbps) saw the highest speeds out of all the markets where the device has launched, reflecting the very high penetration of multi-gigabit fibre in both.
The lack of 320 MHz support does not yet impact N1 performance in the wild
The N1’s performance not only surpasses its Broadcom-based predecessor but also places the iPhone 17 family in a strong competitive position across all Wi-Fi metrics in every region. Notably, Apple’s latest lineup achieved the highest global 10th-percentile download speed at 56.08 Mbps, reinforcing the observation that the N1 is likely to deliver more consistent performance in non-ideal Wi-Fi conditions.
The N1’s apparent handicap on paper, with channel width capped at 160 MHz rather than the 320 MHz that Wi-Fi 7 supports with 6 GHz, does not materially affect performance in real world use for most people. In theory, this cap could halve peak link rates right next to a top tier router, yet the impact is rarely visible outside controlled tests, highlighting the importance of real-world testing and crowdsourced data to reflect the actual end-user experience.
Strong iPhone 17 Performance in North America
Speedtest Intelligence® | Sept 19 – Oct 29, 2025. North America.
This is evident in the iPhone 17 family posting the highest median (416.14 Mbps) and 90th percentile (976.39 Mbps) download speeds of any device in North America, where gains from 320 MHz channels should be most apparent. The most likely explanation is that the installed base of 320 MHz-capable routers remains very small (and our recent shows Wi-Fi 7 adoption itself is still limited), so usage is not yet material enough to move results at the aggregate level.
North American iPhone 17 Speeds Hold Up Without 320 MHz
Speedtest Intelligence® | Sept 19 – Oct 29, 2025. North America.
This may also explain why Apple chose not to add the capability to the N1, even though the performance benefit of 320-MHz-capable silicon is likely to grow as the Wi-Fi ecosystem matures, making it a future-proofing feature for Android flagships that include it.
Google’s Pixel 10 Pro leads on median download speed, Samsung’s Galaxy S25 delivers lowest best-case latency
Beyond the iPhone 17 family, Google’s Pixel 10 Pro also performed strongly on download speed. Likely powered by Broadcom Wi-Fi silicon (consistent with the Pixel 8 and 9 lineage), it achieved the highest global median download speed at 335.33 Mbps during the study period, narrowly ahead of the iPhone 17 family at 329.56 Mbps. In markets such as North America, where Chinese Android brands have limited share, the Pixel 10 Pro also leads in upload performance at both the median and the 90th percentile.
Pixel 10 Pro Leads Global Wi-Fi Download Speeds
Speedtest Intelligence® | Sept 19 – Oct 29, 2025. Global.
Samsung’s Galaxy S25 family, based on Qualcomm’s FastConnect 7900 Wi-Fi silicon integrated with the Snapdragon 8 Elite platform, did not lead outright in any metric at the global level but was positioned in the upper mid-pack across most. Its clearest regional strength was latency, where it delivered the lowest best-case response times in North America (6 ms), Europe (7 ms) and the Gulf (9 ms). It also led in median multi-server latency in Europe (17 ms) and 90th percentile upload speeds in the Gulf (330.80 Mbps).
Xiaomi’s 15T Pro dominates upload performance with MediaTek Wi-Fi silicon
During the study period, the device ranking for upload speed differed markedly from the download ranking, even after controlling for country mix effects (that is, cases where devices skew toward markets with unusually high or low upload speeds). In markets where it has a large installed base, including Europe and Northeast Asia, Xiaomi’s 15T Pro, built on MediaTek Wi-Fi silicon integrated in the Dimensity 9400 (+) platform, showed a commanding lead in upload performance.
During the study period, Xiaomi’s 15T Pro achieved the fastest upload speeds in Europe at every percentile measured (10th, median, 90th) and also led 10th percentile uploads in Northeast Asia. In fiber-rich markets such as France, which are characterized by very high upstream performance and symmetrical line speeds, the 15T Pro was the only device to surpass 100 Mbps at the 10th percentile, 500 Mbps at the median, and 1,000 Mbps at the 90th percentile.
Xiaomi’s 15T Pro Leads on Upload Speed
Speedtest Intelligence® | Sept 19 – Oct 29, 2025. Global.
Beyond upload performance, Xiaomi’s flagship also provided strong performance on multi-server latency, delivering the lowest response times globally at the median (15 ms) and 90th percentile levels (42 ms).
Huawei’s Pura 80 family performs relatively more strongly where 6 GHz is not used
The Pura 80 series is based on a “self-developed chip-level collaboration” for Wi-Fi 7, suggesting, but not confirming, continued use of a HiSilicon solution after the Pura 70’s in-house silicon. If this is the case, Huawei would be the only other manufacturer besides Apple using vertically integrated Wi-Fi silicon across its current flagship lineup.
Critically, however, Huawei’s Wi-Fi 7 implementation in the Pura 80 family lacks 6 GHz support, both on devices sold in China (where 6 GHz is not available for Wi-Fi anyway) and overseas. This limitation significantly impedes performance capability on 6 GHz-capable Wi-Fi networks, especially in crowded environments, where the additional spectrum unlocks major speed gains on devices that can take advantage of it.
Huawei's Pura 80 Performs Better on Non-6 GHz Wi-Fi
Speedtest Intelligence® | Sept 19 – Oct 29, 2025. Southeast Asia.
The lack of 6 GHz support is particularly evident at the 90th percentile, where the Pura 80 family trailed all other devices in Southeast Asia, the region with the largest observed install base for the device, posting download speeds of 541.33 Mbps that were more than 39% below the top performing Oppo Find X8 Pro there. This lag also extended to median download speeds in the same region, where the Pura 80 family again trailed all other devices.
Notwithstanding this disadvantage, the Pura 80 was competitive on some metrics, including upload performance on access points lacking Wi-Fi 6E and Wi-Fi 7 (which do not benefit from 6 GHz access). On Wi-Fi 6 connections, Huawei’s flagship delivered the second-fastest upload speeds at the 90th percentile (603.61 Mbps) in Southeast Asia against Android flagships.
Wi-Fi 7 and 6 GHz propel flagships to new performance levels, but benefits remain fragmented
Although Wi-Fi outcomes vary by device, even between models using the same silicon because factors like hardware and software integration and chassis tuning affect results, and although they also vary by region, the commonality is a step-change in performance on flagship devices enabled by newer standards such as Wi-Fi 7 and access to the 6 GHz band.
North American Flagship Users Spend More Time on 6 GHz Wi-Fi
Speedtest Intelligence® | Sept 19 – Oct 29, 2025. Samsung Galaxy S25 Family.
On modern access points and devices with Wi-Fi 7-capable silicon, users can take advantage of newer features like Multi-Link Operation (MLO), which enables the use of multiple Wi-Fi bands at the same time (similar to carrier aggregation with cellular).
Flagship Devices See Higher Speeds on Newer Wi-Fi Standards
Speedtest Intelligence® | Sept 19 – Oct 29, 2025. Global.
These upgrades are translating into tangible gains, with Wi-Fi 7 delivering roughly double the median download speeds of Wi-Fi 6 on the same flagship Android devices included in this study (uplift ranging from +74% to +108% depending on device family). The step from Wi-Fi 5 to Wi-Fi 6 delivered a similar uplift on these devices (uplift ranging from +72% to +123%). Similarly, median download speeds on flagship devices connected to 6 GHz were at least 77% faster than 5 GHz.
Flagship Devices Perform Better on Higher Wi-Fi Bands
Speedtest Intelligence® | Sept 19 – Oct 29, 2025. Global.
The diffusion of these benefits in the real-world, however, is still at an early stage and regionally fragmented. For instance, while over 20% of Speedtest samples conducted on the Galaxy S25 family in North America originated on the 6 GHz band during the study period, only about 5% of samples in Europe and Northeast Asia and 1.7% in the Gulf region were based on 6 GHz.
Ookla retains ownership of this article including all of the intellectual property rights, data, content graphs and analysis. This article may not be quoted, reproduced, distributed or published for any commercial purpose without prior consent. Members of the press and others using the findings in this article for non-commercial purposes are welcome to publicly share and link to report information with attribution to Ookla.
Luke Kehoe leads Ookla’s research and thought leadership efforts in Europe.
An electronic engineering alumnus of University College Dublin, Luke has extensive experience collaborating with mobile operators, telecoms vendors, and government agencies in research and advisory roles across Europe. He has contributed to internationally recognised thought leadership publications in areas such as 5G, IoT, open RAN, and edge computing, working with prestigious organisations like the Telecom Infra Project and the World Economic Forum.
Single points of logical failure topple even the most hardened cloud infrastructure, crippling today’s highly concentrated internet ecosystem.
Editor’s note: This article was updated on October 22 following a 48-hour review period related to the AWS outage, reflecting an upward revision of the total Downdetector user report volume from 16M+ to 17M+.
An Amazon Web Services (AWS) disruption on October 20, 2025, centered on its “US‑EAST‑1” cloud region triggered a wave of failures across consumer apps, finance, government portals and parts of Amazon’s own services. Downdetector® recorded 17M+ user reports (+970% increase on average daily baseline) and disruptions at over 3,500 companies across more than 60 countries, placing this among the largest internet outages on record for Downdetector.
Key Takeaways:
This outage had exceptional global reach and deep cross-sector impact. Downdetector captured 17M+ outage reports globally across 60+ countries, with the US (>6.3M) and UK (>1.5M) leading outage volumes. Services with the most reports included Snapchat (~3M), Roblox (~716k) and Amazon retail (~698k), and spanned everything from banking to gaming services.
Rapid cascade and phased restoration. Thefirst outage spikes appeared around 06:50–07:00 UTC. AWS identified DNS resolution issues affecting DynamoDB endpoints in US‑EAST‑1 and reported mitigation by 09:24 UTC, with full normalization later in the day as downstream services cleared backlogs on a phased basis. Outage reports underwent a second surge in the late afternoon (UTC) as U.S. users awoke to disruptions.
Not a one-off event. The outage echoes recent systemic failures, including Meta’s 2021 BGP/DNS issue, Fastly and Akamai CDN outages, the 2024 CrowdStrike update failure, the 2025 Cloudflare-AWS interconnect incident and the recent Google Cloud outage, revealing single points of failure in shared infrastructure.
Wake-up call for critical infrastructure. The lesson is concentration risk (or overreliance on a single point of failure). The service layer is now tightly coupled to a handful of cloud regions and managed services. The way forward is not zero failure but contained failure, achieved through multi-region designs, dependency diversity, and disciplined incident readiness, with regulatory oversight that moves toward treating the cloud as systemic components of national and economic resilience.
The “blast radius” reached far beyond Virginia, where the affected AWS US-EAST-1 is located
Downdetector captured 17M+ global user reports from 00:00 UTC on Oct 20 to 09:15 UTC on Oct 21, with 3,500+ of the companies Downdetector tracks seeing elevated disruptions and 19 still ongoing the following morning. Country volumes were led by the US (6.3M+), UK (1.5M+), Germany (774k), Netherlands (737k) and Brazil (589k).
The heaviest‑hit services by report count included Snapchat (~3M), AWS itself (~2.5M), Roblox (~716k), Amazon retail (~698k), Reddit (~397k), Ring (~357k) and Instructure (~265k). The UK alone generated more than 1.5M reports, far exceeding a typical day’s ~1M global baseline across all markets, highlighting both the unique intensity and breadth of this event.
Analysis of sectoral outcomes in Downdetector reports reveal impacts spanned social/gaming (Snapchat, Fortnite, Roblox), finance (e.g., UK banks like Lloyds and Halifax), public services (HMRC), smart home (Ring, Alexa), and education/work tools (Instructure, Zoom). Outage peaks and troughs varied by time zone, with European volumes rising first as workplaces came online and a second lift as North America woke up later in the day:
~06:49 UTC (Oct 20): First user reports and AWS status signals line up. Downdetector registered sharp spikes shortly after 06:56 UTC on US‑EAST‑1‑linked services. Within two hours of outage commencement, over 4M outage reports were submitted.
~09:24 UTC: AWS says the core fault, involving DNS resolution issues for regional DynamoDB endpoints in US‑EAST‑1, was mitigated.
Remainder of the day: Dependent services recovered at different speeds as retries, queues and caches drained. Major outlets tracked staged restoration through the afternoon and evening local time, with Downdetector reports surging past 6M in the U.S. as users came online.
This pattern, reflecting a relatively short underlying cloud incident based on a common denominator (AWS US-EAST-1 concentration) with longer downstream normalization, is customary when a foundational component (in this case DNS to a regional database endpoint) sits behind many higher‑level services and microservices.
Regional concentration and tight coupling of managed services amplified outage impact
The affected US‑EAST‑1 is AWS’s oldest and most heavily used hub. Regional concentration means even global apps often anchor identity, state or metadata flows there. When a regional dependency fails as was the case in this event, impacts propagate worldwide because many “global” stacks route through Virginia at some point.
Modern apps chain together managed services like storage, queues, and serverless functions. If DNS cannot reliably resolve a critical endpoint (for example, the DynamoDB API involved here), errors cascade through upstream APIs and cause visible failures in apps users do not associate with AWS. That is precisely what Downdetector recorded across Snapchat, Roblox, Signal, Ring, HMRC, and others.
Another complicating factor in this outage was authentication. Problems with DynamoDB also hit IAM (authentication), which handles sign in and permissions. Early on, some teams could not log in to the AWS console. Where teams cannot sign in to the tools that change settings, move traffic, or restart services, it is very difficult to apply fixes, so recovery slows even after core systems start to come back.
Even after the provider (AWS in this case) fully mitigated the issue, retries, timeouts and message backlogs took time to clear. Teams often throttle restarts to protect back ends, so user-visible recovery lags the provider’s green status. The afternoon and evening recovery curve observed in Downdetector reports, first in Europe and then in the U.S., matches this pattern.
Companies should plan for region failure and practise graceful slowdowns during outages
For companies dependent on these platforms, a practical response to an outage of this magnitude begins with designing for failure and assuming a whole cloud region can go down. This means not relying on a single region (e.g., US-EAST-1) for critical systems and instead running services across multiple regions (known as active-active) or keeping a lightweight standby (“pilot light”) that can be promoted quickly. For highly mission-critical services, the use of a multi-cloud setup can improve availability during provider-wide incidents, but this is not practical for many companies due to the costs of duplication and additional complexity.
Similarly, companies should plan for graceful slowdowns, not just total outages. This means using circuit breakers and feature flags to turn off non-essential features (like media uploads or recommendations) so that core flows such as sign-in, search, or checkout stay up. Practicing the act of “failing safely” is important, often achieved through running game days that simulate DNS, database, and authentication outages. Investing in measuring time to detect, time to fail over, and understanding how clear customer communications are is also important.
In real incidents like this, it is important that companies compare their own telemetry with public signals (e.g., Downdetector) to understand if the issue is provider-wide. When the facts are established, early communication with customers via status pages or in-app notifications is critical to reduce support load and protect trust.
Policymakers should treat cloud infrastructure as systemic components of national digital resilience
This outage again shows that cloud platforms are systemic infrastructure, characterized by a massive blast radius when a single point of logical failure emerges. It highlights the limitations of investing in enhancing physical resilience and redundancy alone (e.g., multi-day backup power on-site) if there is a fault elsewhere in the infrastructure stack on which everything else depends and from which failures cascade. Improving outcomes requires dismantling these single points of failure through diversification across each layer.
At a policy level, governments are starting to recognise this systemic risk and are beginning to adopt a more muscular approach to oversight. The EU’s flagship Digital Operational Resilience Act (DORA) introduces EU-level oversight of critical ICT third-party providers, while the UK’s Critical Third Parties regime does the same for finance.
Together, these policy developments aim to create a stronger toolkit, based on dependency mapping, stress tests, incident reporting discipline, and minimum post-event transparency, that will likely (and should) in time extend beyond financial services to other essential sectors (e.g., health, transport, and government). Importantly, proponents argue that adoption of these approaches would improve societal resilience without unduly micromanaging architectural choices.
For businesses, Downdetector provides access to dashboards that deliver early alerts, enable outage correlation, and allow for direct communication with users, ensuring a proactive approach to incident management. Learn how you can leverage Downdetector to be better prepared for outages, or reach out to schedule a demo.
Ookla retains ownership of this article including all of the intellectual property rights, data, content graphs and analysis. This article may not be quoted, reproduced, distributed or published for any commercial purpose without prior consent. Members of the press and others using the findings in this article for non-commercial purposes are welcome to publicly share and link to report information with attribution to Ookla.
Luke Kehoe leads Ookla’s research and thought leadership efforts in Europe.
An electronic engineering alumnus of University College Dublin, Luke has extensive experience collaborating with mobile operators, telecoms vendors, and government agencies in research and advisory roles across Europe. He has contributed to internationally recognised thought leadership publications in areas such as 5G, IoT, open RAN, and edge computing, working with prestigious organisations like the Telecom Infra Project and the World Economic Forum.
Reliability-led differentiation hinges on low-band reach, mid-band density, and diversified core interconnects
Portugal’s mobile operators are on the defensive as DIGI scales its low-cost, flexible offers. With aggressive pricing across converged bundles that undercut the market by a wide margin, the Romanian disruptor is compressing margins at the value end, driving elevated churn and forcing operators to raise retention opex through cheaper flanker brands and shorter contract terms.
By choosing a rapid greenfield own-build rather than a national roaming agreement, DIGI has kept competition price-led rather than coverage-led. This has elevated network reliability as the established operators’ principal strategic moat and encouraged them to leverage their more mature infrastructure across Portugal as a core differentiator. The Iberian grid blackout earlier this year underscored the point: limited power autonomy and weak geo-redundancy in a greenfield network stack contributed to markedly poorer resilience outcomes, as revealed by first-of-its-kind analysis of background signal scans using Ookla® data.
The established operators’ are now seeking to copper-fasten their resilience and reliability credentials through heavy capital spending on network modernization, focused on diversifying spectrum use for more capable carrier aggregation, commercializing 5G Standalone (SA), and densifying the grid footprint.
To quantify how the network reliability race is unfolding in Portugal’s mobile market, we independently measured performance using RootMetrics’ controlled methodology on the latest Samsung flagship handsets. Testing in 1H 2025 covered indoor and outdoor locations across Lisbon, Porto, Madeira, the Azores, and an extensive national route. Combining walk and drive testing in high-usage areas, we covered more than 9,500 km and collected nearly 110,000 samples, including 146 indoor sites. The methodology is designed to mirror real-world network performance.
Key Takeaways:
MEO and Vodafone lead Portugal in network reliability. MEO prioritizes a wide, contiguous mid-band layer almost everywhere, underpinned by extensive low-band coverage. This yields the highest access and task success on the national route and the fastest call setup times in Portugal, as well as driving its Best 5G award based on Speedtest® data in 1H 2025. Vodafone, for its part, tops access latency and video task reliability in Lisbon and Porto, helped by advanced use of carrier aggregation, a deep low-band layer, and well-peered core paths.
NOS reaps a first-mover advantage from its expanding 5G SA footprint. The operator’s 5G SA deployments remain heavily urban-weighted, with 56% of Lisbon samples on the new core architecture versus 9% on the national route. Its strategy is to deploy 5G SA where there is dense mid-band (3.5 GHz) and carrier aggregation, and to fall back to NSA and low-band spectrum (700 MHz) elsewhere. The latency benefits of moving to the 5G core are validated by NOS’s leading video start times and access latency nationally.
DIGI’s network is mid-band-centric on narrow spectrum with limited carrier aggregation. The early-stage profile of the rollout is evident in its disproportionate reliance on the 2.6 GHz band and the absence of a low-band layer, which constrains coverage reach and increases handover exposure, weighing on network reliability. It does, however, deliver strong 5th percentile performance and national call setup times that are competitive with leader MEO, yet its time on 5G remains materially below the incumbents in every geography tested.
Regional network performance disparities persist in Portugal. Performance is poorer on Madeira and in the Azores because radio grids are sparser and rely more on low-band spectrum to span difficult terrain, so devices see fewer mid-band carriers and less carrier aggregation. Lower time on 5G leads to more frequent moves between network generations, increasing the risk of late handovers and task failures. Core breakout and peering are also less distributed, so traffic often routes via longer paths to mainland gateways, adding delay at busy times.
Network reliability challenges operators to optimize networks across RAN and core layers
MEO and Vodafone Lead in Network Reliability across Portugal
RootMetrics® | 1H 2025
To compare, in a scientifically robust way, how operators’ network investments translate into reliability in Portugal, RootMetrics’ controlled testing asks a simple question: when a user starts a task, does it complete without failing? Tens of thousands of “connect and complete” tests spanning calls, data uploads and downloads, and texts are conducted across varied routes and locations, then aggregated into a single Reliability score. The score is weighted toward data (75%), including calls (20%) and texts (5%) to reflect today’s real-world usage patterns.
The methodology rewards successful starts and uninterrupted completion and penalizes blocks, drops, and timeouts. Since each test follows the full path from device to radio to core to service edge, the results reflect end-to-end robustness rather than any single parameter:
Call Reliability (20% overall weight): This measures voice connection stability. It assigns more weighting to blocking (user presses call and the network refuses or never sets it up) over dropping (the call starts but ends unexpectedly), because initial failures tend to disrupt user intent more profoundly.
Blocking often rises during load spikes when signalling or media resources are exhausted at busy venues or during emergencies. Dropping usually increases with poor radio conditions and handover problems, for example low SINR and coverage gaps in rural areas.
Data Reliability (75% overall weight): This measures whether devices can establish a secure, usable data path (access success) and complete common transfers (task success) without stalls or timeouts. It covers both download and upload under light tasks such as webpage loads and heavier tasks such as file transfers, rewarding successful setup and uninterrupted completion and penalizing setup failures, timeouts, and mid-flow resets.
Even in cases where users see full signal bars on their device, data reliability components like task success can decline due to factors like packet loss and TCP resets (e.g., at a busy stadium) or poor mid-transfer handover (e.g., while on a high-speed train).
Text Reliability: (5% overall weight): This measures the ability to send and receive texts consistently, both within the same operator and across operators.
While text performance is usually robust, it can fail during incidents that span interconnect outages, misrouted numbering, spurious anti-spam filtering or queue backlogs.
Analysis of the controlled testing data in Portugal reveals the most reliable networks consistently combine adequate mid-band capacity to keep the median user out of congestion, meaningful use of low-band spectrum to lift access and task success at the cell edge and indoors, and tight integration between core and radio layers that keeps call setup fast and access latency low.
DIGI’s greenfield buildout gathers pace as established operators focus on RAN modernization and grid densification
DIGI’s commercial launch in Portugal in November 2024 has triggered a flurry of defensive moves by the country’s established operators. Confronted with a sharp rise in churn and renewed ARPU pressure, they have pushed more aggressive retention offers and leaned on low-cost flanker brands. DIGI’s disruptive model has reset reference pricing for entry-level mobile and convergent bundles. Even where rivals have not matched the headline tariffs, their flanker brands now cluster around €8 to €15 for 100 GB tiers to compete with DIGI’s eye-catching plans from as little as €4 for 50 GB, as well as seeking to limit cannibalization of their core brands.
Analysis of ANACOM data on subscriptions and site counts show DIGI’s rise has been exceptional. By Q1 2025 it already held over 3% of mobile internet subscriptions and rolled out 2,385 5G sites. With no national roaming and no low-band spectrum, it has concentrated capex on rapid coverage, densifying urban and suburban grids to compete on performance and offset the less favourable propagation of its mid-band holdings.
Vodafone and NOS Feature Largest 5G Site Footprint in Portugal
ANACOM | Q1 2025
The maturity of the established operators’ infrastructure, and the imperative to defend their network performance moat, puts them at a very different point in the 5G investment cycle to DIGI. MEO, for example, is partway through a multi-year RAN modernization and swap to Nokia, diverting capex into replacing legacy equipment and lifting performance and energy efficiency across its footprint.
NOS and Vodafone, which share mobile infrastructure under a MORAN arrangement (no spectrum sharing) in rural and interior areas to lower site costs and deepen rural coverage, have built leading 5G footprints after earlier, front-loaded expansion capex. NOS in particular has sought to differentiate through early 5G SA commercialization, marketing its Nokia-supplied 5G core as “5G+” and touting higher uplink speeds, lower latency, and better device battery life. MEO has since followed suit and deployed 5G SA atop its 3.5 GHz sites.
Mid-band depth and a 700 MHz underlay puts MEO at the frontier of network access and task success nationally
MEO’s relatively smaller 5G site footprint has driven a focus on efficient spectrum use over sheer volume as it ramps up site expansion (30% increase in 5G site count over the last year, the fastest rate of any operator in the period). Its balanced strategy leans on mid-band capacity with low-band for coverage extension. MEO recorded the highest share of 3.5 GHz (n78) usage in every geography in testing, operating a wide 90 MHz channel nationally. The contiguous mid-band block is the most extensively deployed in Portugal, with more than two-thirds of national route samples using it, 76-80% the islands, and 88% in Lisbon.
Outside dense urban areas like Lisbon and Porto, the operator makes extensive use of the 700 MHz (n28) band as a coverage layer. Nationally, roughly one-third of 5G samples were on 5 MHz of 700 MHz, while Lisbon’s 700 MHz share fell below 10% and the islands sat between 20 and 22%. This strategy maximizes mid-band time-on-air and keeps modulation high for the median user while preserving coverage probability at the edge, which reduces access failures and video stalls.
Because its low-band channel is narrower than rivals (Vodafone and NOS use 10 MHz), its overall 5G site footprint is smaller, and its spectrum mix beyond 700 MHz and 3.5 GHz is limited, with little carrier aggregation and sparse use of 2.1 GHz, devices on MEO’s network spend less time on 5G than those on Vodafone and NOS.
MEO Leads on Call Setup Times, while Vodafone Features Lowest Call Drop Rate
RootMetrics® | 1H 2025 – National Route
Despite this, MEO co-led reliability on the national route, delivering the highest access and task success and the fastest call setup at about 1.8 seconds. Madeira showed the same pattern, with MEO leading on access and task success, best call setup, and top overall reliability. In Lisbon and Porto, MEO was neck and neck with Vodafone on access and task success, and slightly ahead on call drop rate in the capital. The takeaway is that a higher share of time on 5G (enjoyed by Vodafone and NOS thanks to a larger 5G site footprint) and greater use of carrier aggregation does not automatically translate into a more consistent network experience.
Intensive use of carrier aggregation drives spectrum diversity across Vodafone’s site footprint
By Q4 2024, Vodafone’s extensive 5G site footprint placed it alongside NOS for time spent on 5G. In the testing, 84% of Vodafone samples were based on 5G nationally, compared with 66% for MEO and 27% for DIGI. Vodafone’s network profile closely mirrors NOS’s even in passively shared urban areas outside the regions where the pair engage in active sharing, with both showing intensive use of carrier aggregation. Vodafone, unlike NOS, still lacks a 5G SA footprint.
The operator combines its contiguous 700 MHz (10 MHz) and 3.5 GHz (90 MHz) holdings through dual-carrier aggregation (2CC), creating a potent 100 MHz aggregated capacity layer. The 3.5 GHz width matches MEO’s but is 10 MHz narrower than NOS’s. Across the national route, 2CC featured in just over one fifth of 5G samples, rising to a remarkable two-thirds of all samples in Lisbon and Porto. On the islands, where the grid is sparser, testing shows that Vodafone favours a balanced split between 700 MHz and 3.5 GHz for 5G with limited use of carrier aggregation.
Different 5G Spectrum Playbooks: Meo Prioritizes 3.5 GHz, NOS and Vodafone Leverage Carrier Aggregation
RootMetrics® | 1H 2025 (Observed Spectrum Use and Bandwidth, National Route)
Vodafone’s advanced use of 2CC lifts both peak and median performance, and it performs the strongest of all operators at the tail (5th percentile) nationally. This points to fair airtime sharing through effective resource scheduling and strong link adaptation that consistently selects the best modulation and coding, so performance at the cell edge or in busy cells holds up even when carrier aggregation is limited.
These outcomes translate into strong reliability. Vodafone jointly led overall network reliability with MEO in testing and led outright in the Azores, recording the lowest dropped call rate on the national route. In Lisbon and Porto, Vodafone delivered the best access latency and ranked first or joint first for video reliability, demonstrating that optimizations beyond deploying an SA core, such as short core paths and close CDN proximity, can drive strong results.
Early 5G SA underpins NOS’s lead in network responsiveness
Thetesting revealed that NOS features the most advanced network configuration in Portugal, combining a large and growing 5G SA footprint with the broadest range of carrier aggregation combinations. It also operates the largest 5G site grid, nearing 4,800 sites by Q4 2024, which put it neck and neck with Vodafone for time spent on 5G nationally at 84%.
Like Vodafone, NOS pairs a wide mid-band allocation (100 MHz at 3.5 GHz) with low-band spectrum (10 MHz at 700 MHz) and applies aggressive carrier aggregation across both NSA and SA. In Lisbon, 2CC was observed on 59% of samples, delivering the highest capacity configuration of any operator in Portugal at 110 MHz of aggregated spectrum, thanks to the wider mid-band block. On the national route, 2CC appeared in 20% of samples. Notwithstanding this, NOS showed a notably lower use of the 3.5 GHz band on the national route, at 32% of samples, compared with Vodafone at 37% and MEO at 67.6%.
NOS’s first mover advantage in 5G SA is translating into superior access responsiveness and faster video starts in cities. In Lisbon, 94% of samples were on 5G and 56% on SA, delivering the quickest video start at about 530 ms; in Porto it again posted the quickest start at about 540 ms. Outside dense areas, where SA represents only 9% of 5G usage on the national route, NOS still led these metrics, indicating disciplined core routing with clean CDN peering (traffic exiting core close to the user) and a fast, low-latency scheduler that delivers a quick first byte regardless of SA.
Low-band absence drives DIGI to maximize mid-band buildout
DIGI’s greenfield buildout has had to do more with less. Lacking 700 MHz spectrum to anchor a wide-area 5G layer, it has leaned heavily on mid-band assets, which drives more frequent 4G fallback and patchier 5G access indoors and in rural areas. In testing, only 27% of samples on the national route were on DIGI’s 5G network (and only 16% in the Azores and 21% on Madeira), rising to 48% in Lisbon, highlighting the urban skew of its initial rollout.
The operator’s mid-band 5G deployments are concentrated on relatively narrow TDD carriers rather than a single wide 3.5 GHz channel, reflecting the limits of a non-contiguous assignment. DIGI’s two 3.5 GHz carriers are separated by 40 MHz: it initially held one 40 MHz block from the 2021 award and received another from NOWO earlier this year, while Dense Air retains spectrum between them. Carrier aggregation can soften some drawbacks of a split 3.5 GHz allocation, but there remains an inherent efficiency and device support gap for NR intra-band CA versus a single wide carrier, which is less favourable from an RF engineering perspective.
NOS and Vodafone Boast Advanced Carrier Aggregation Depth in Lisbon
RootMetrics® | 1H 2025 (Observed Network Architecture Share, Lisbon)
In Lisbon, about 98% of DIGI’s 5G samples were based on a 20 MHz block of 2.6 GHz (n41) spectrum. On the national route, its mix split more evenly across 3.5 GHz (mostly a 40 MHz n78 configuration, less than half the width of other operators’ n78 assignments) and 2.6 GHz (20 MHz in each of the n38 and n41 bands). It has deployed the 3.5 GHz band more extensively in Porto than in Lisbon, with 37% of time spent on that spectrum in the former.
The nascent nature of DIGI’s greenfield buildout is reflected in its trailing peers on overall network reliability in testing, with lower network access and task success rates across all regions. While it already shows strong performance in call setup time (joint-best with MEO), poor outcomes in call drop rate (behind peers) indicate that factors such as a sparse 5G anchor and handover instability continue to pull down performance.
Conclusion: No one-size-fits-all in the race for network reliability in Portugal
The diversity in spectrum assignments (frequencies, bandwidths, and contiguity), subscriber base profiles (size, location, and traffic demand), and network investment cycles (greenfield buildout, vendor swap, or simple RAN refresh) means Portugal’s operators must pull different levers in their pursuit of competitive differentiation through network reliability. Despite these differences, all operators share an emphasis on low-band coverage for reach, mid-band capacity for depth, diverse and robust core interconnections, and strategic CDN placement
MEO and Vodafone’s lead in overall network reliability stems from distinctly different approaches, with MEO leveraging aggressive deployment of a wide 3.5 GHz block nearly everywhere, while Vodafone applies advanced carrier aggregation selectively, highlighting that no single strategy fits all in the quest for superior reliability. Likewise, NOS’s early rollout of 5G SA is enhancing latency-sensitive performance, and DIGI is delivering strong initial results in metrics like call setup time.
Novo concorrente impulsiona uma corrida à fiabilidade de rede no mercado móvel português
A diferenciação baseada na fiabilidade assenta no alcance em faixa baixa, na densidade em faixa média e em interligações diversificadas no core
Os operadores móveis em Portugal encontram-se na defensiva à medida que a DIGI expande as suas ofertas de baixo custo e elevada flexibilidade. Com uma política de preços agressiva em pacotes convergentes que reduzem significativamente o valor médio de mercado, o operador romeno está a comprimir as margens no segmento de valor, a aumentar a rotatividade de clientes e a obrigar os operadores a reforçar os gastos operacionais de retenção através de marcas secundárias mais baratas e contratos de menor duração.
Ao optar por uma rápida construção de rede própria em vez de um acordo de roaming nacional, a DIGI manteve a concorrência centrada no preço e não na cobertura. Isso elevou a fiabilidade da rede ao estatuto de principal trunfo estratégico dos operadores estabelecidos, incentivando-os a tirar partido das suas infraestruturas mais maduras em todo o país como elemento diferenciador central. O apagão da rede ibérica ocorrido no início deste ano veio sublinhar esta realidade: a autonomia energética limitada e a fraca redundância geográfica de uma rede recém-construída contribuíram para resultados de resiliência significativamente inferiores, conforme demonstrado por uma análise inédita de varrimentos de sinal de fundo com base em dados da Ookla®.
Os operadores estabelecidos procuram agora cimentar as suas credenciais de resiliência e fiabilidade através de forte investimento de capital na modernização da rede, com foco na diversificação do uso de espectro para uma agregação de portadoras mais capaz, na comercialização do 5G Standalone (SA) e na densificação da malha de cobertura..Para quantificar como está a evoluir a corrida à fiabilidade de rede no mercado móvel português, medimos de forma independente o desempenho das redes utilizando a metodologia controlada da RootMetrics, com os mais recentes modelos topo de gama da Samsung.. Os testes realizados no primeiro semestre de 2025 abrangeram locais interiores e exteriores em Lisboa, Porto, Madeira, Açores e uma extensa rota nacional. Combinando testes em caminhada e condução em zonas de elevada utilização, cobrimos mais de 9.500 km e recolhemos cerca de 110.000 amostras, incluindo 146 locais interiores. A metodologia foi concebida para reproduzir o desempenho real das redes.
Principais conclusões:
A MEO e a Vodafone lideram em fiabilidade de rede em Portugal. A MEO privilegia uma camada de banda média ampla e contínua quase em todo o território, sustentada por uma cobertura extensiva em banda baixa. Este equilíbrio proporciona as melhores taxas de acesso e de sucesso de tarefas na rota nacional, os tempos de estabelecimento de chamadas mais rápidos do país e fundamenta o seu prémio de “Melhor 5G” com base em dados Speedtest no primeiro semestre de 2025.
A Vodafone, por sua vez, lidera em latência de acesso e fiabilidade em tarefas de vídeo em Lisboa e no Porto, beneficiando de um uso avançado de agregação de portadoras, de uma camada profunda de banda baixa e de percursos de núcleo bem interligados.
A NOS colhe uma vantagem de pioneiro com a expansão da sua rede 5G SA. As implementações de 5G SA do operador permanecem fortemente concentradas em meio urbano, com 56% das amostras em Lisboa já sobre a nova arquitetura de núcleo, contra 9% na rota nacional. A estratégia passa por ativar o 5G SA onde existe densidade de banda média (3,5 GHz) e agregação de portadoras, recorrendo ao 5G NSA e à banda baixa (700 MHz) nas restantes zonas. Os ganhos de latência decorrentes da migração para o novo núcleo são comprovados pelos melhores tempos de início de vídeo e menor latência de acesso registados pela NOS a nível nacional.
A rede da DIGI é centrada na banda média, com espectro limitado e agregação reduzida. O caráter inicial do seu desenvolvimento é evidente na dependência desproporcionada da faixa dos 2,6 GHz e na ausência de uma camada de banda baixa, o que restringe o alcance de cobertura e aumenta a exposição a transições entre células, penalizando a fiabilidade da rede. Ainda assim, apresenta um desempenho sólido no quinto percentil e tempos de estabelecimento de chamadas a nível nacional competitivos face à líder MEO, embora o tempo em 5G permaneça significativamente inferior ao dos operadores incumbentes em todas as geografias testadas.
Persistem disparidades regionais de desempenho de rede em Portugal. O desempenho é inferior na Madeira e nos Açores devido a redes rádio mais dispersas e maior dependência de espectro de banda baixa para cobrir terrenos difíceis, o que resulta em menor presença de portadoras de banda média e menos agregação. O menor tempo em 5G conduz a transições mais frequentes entre gerações de rede, aumentando o risco de falhas de transição e de tarefas. A distribuição limitada dos pontos de interligação do núcleo também implica que o tráfego seja frequentemente encaminhado por percursos mais longos até gateways no continente, o que acarreta atrasos em períodos de maior utilização.
A fiabilidade de rede obriga os operadores a otimizar as redes nas camadas RAN e Núcle
MEO and Vodafone Lead in Network Reliability across Portugal
RootMetrics® | 1H 2025
Para comparar, de forma cientificamente rigorosa, de que modo os investimentos dos operadores se traduzem em fiabilidade em Portugal, a metodologia controlada da RootMetrics coloca uma questão simples: quando o utilizador inicia uma tarefa, esta conclui-se sem falhar?
São realizados dezenas de milhares de testes “conectar e concluir”, abrangendo chamadas, transferências de dados e mensagens de texto em diferentes percursos e locais, agregados depois num único índice de Fiabilidade. Este índice atribui maior peso aos dados (75%), incluindo chamadas (20%) e mensagens (5%), refletindo os padrões de utilização atuais.
A metodologia recompensa inícios bem-sucedidos e conclusões sem interrupções, penalizando bloqueios, quedas e falhas de tempo limite. Como cada teste cobre o percurso completo — do dispositivo à rádio, ao núcleo e até à extremidade do serviço —, os resultados refletem a robustez de ponta a ponta e não apenas um parâmetro isolado.
Fiabilidade de Chamadas (peso de 20% no total): Mede a estabilidade das ligações de voz, atribuindo maior peso aos bloqueios (quando o utilizador tenta ligar e a rede não responde ou não estabelece a chamada) do que às quedas (quando a chamada inicia, mas termina inesperadamente), dado que as falhas iniciais têm um impacto mais disruptivo na experiência do utilizador.
Os bloqueios tendem a aumentar durante picos de carga, quando os recursos de sinalização ou de media se esgotam em locais muito movimentados ou durante emergências. As quedas, por sua vez, são mais frequentes em zonas com más condições rádio ou falhas de transição entre células, como em áreas rurais.
Fiabilidade de Dados (peso de 75% no total): Avalia se os dispositivos conseguem estabelecer uma ligação de dados segura e utilizável (sucesso de acesso) e concluir transferências comuns (sucesso de tarefa) sem interrupções ou falhas. Inclui tanto downloads como uploads, desde tarefas leves (carregamento de páginas) até transferências mais exigentes, recompensando a conclusão sem falhas e penalizando falhas de configuração, interrupções ou reinícios.
Mesmo quando os utilizadores veem barras de sinal completas, a fiabilidade pode degradar-se devido a perda de pacotes ou reinícios de TCP (por exemplo, num estádio cheio) ou a falhas de transição durante transferências em movimento (como em comboios de alta velocidade).
Fiabilidade de SMS (peso de 5% no total): Mede a capacidade de enviar e receber mensagens de texto de forma consistente, tanto dentro da mesma rede como entre operadores diferentes.
Embora o desempenho neste indicador seja geralmente robusto, podem ocorrer falhas em casos de interrupções de interligação, erros de encaminhamento numérico, filtros anti-spam indevidos ou congestionamento de filas.
A análise dos dados de testes controlados em Portugal revela que as redes mais fiáveis combinam, de forma consistente, capacidade adequada em banda média para evitar congestionamentos, utilização significativa de espectro de banda baixa para melhorar o acesso e o sucesso de tarefas nas extremidades de célula e em interiores, e integração apertada entre as camadas núcleo e rádio, garantindo tempos rápidos de estabelecimento de chamadas e latência reduzida no acesso.
A expansão da rede própria da DIGI ganha ritmo enquanto os operadores estabelecidos apostam na modernização da RAN e na densificação da rede
O lançamento comercial da DIGI em Portugal, em novembro de 2024, desencadeou uma série de movimentos defensivos por parte dos operadores móveis estabelecidos. Perante um aumento acentuado da rotatividade de clientes e uma renovada pressão sobre o ARPU, estes responderam com ofertas de retenção mais agressivas e maior dependência de marcas secundárias de baixo custo. O modelo disruptivo da DIGI redefiniu a referência de preços para os pacotes móveis e convergentes de entrada. Mesmo onde os concorrentes não igualaram as tarifas de destaque, as suas marcas flanqueadoras posicionam-se agora entre os 8 e 15 euros para planos de 100 GB, de modo a competir com as propostas atrativas da DIGI — que começam nos 4 euros por 50 GB —, procurando simultaneamente limitar a canibalização das suas marcas principais.
A análise dos dados da ANACOM sobre assinaturas e número de sites mostra que a ascensão da DIGI tem sido excecional. No primeiro trimestre de 2025, já detinha mais de 3% das subscrições de internet móvel e havia implantado 2.385 sites 5G. Sem roaming nacional e sem espectro em banda baixa, concentrou o investimento de capital em cobertura rápida, densificando as redes urbanas e suburbanas para competir em desempenho e compensar a menor propagação das suas frequências em banda média.
Vodafone and NOS Feature Largest 5G Site Footprint in Portugal
ANACOM | Q1 2025
A maturidade das infraestruturas dos operadores estabelecidos e a necessidade de defenderem a sua vantagem em desempenho de rede colocam-nos numa fase muito distinta do ciclo de investimento em 5G face à DIGI. A MEO, por exemplo, encontra-se a meio de um processo plurianual de modernização da RAN e de substituição de equipamentos pela Nokia, canalizando investimentos para a atualização tecnológica e a melhoria da eficiência energética em toda a sua rede.
A NOS e a Vodafone, que partilham infraestrutura móvel sob um acordo MORAN (sem partilha de espectro) em áreas rurais e do interior para reduzir custos e reforçar a cobertura, consolidaram as maiores redes 5G após um investimento inicial intensivo. A NOS, em particular, procurou diferenciar-se através da comercialização antecipada do 5G Standalone (SA), promovendo o seu núcleo 5G fornecido pela Nokia sob a designação “5G+”, destacando velocidades de upload mais elevadas, menor latência e maior autonomia de bateria nos dispositivos. A MEO seguiu o mesmo caminho e implementou igualmente o 5G SA nas suas antenas de 3,5 GHz.
Densidade de cobertura de banda média e utilização da banda dos 700 MHz colocam a MEO na linha da frente do acesso e sucesso de operações
A NOS e a Vodafone, que partilham infraestrutura móvel sob um acordo MORAN (sem partilha de espectro) em áreas rurais e do interior para reduzir custos e reforçar a cobertura, consolidaram as maiores redes 5G após um investimento inicial intensivo. A NOS, em particular, procurou diferenciar-se através da comercialização antecipada do 5G Standalone (SA), promovendo o seu núcleo 5G fornecido pela Nokia sob a designação “5G+”, destacando velocidades de upload mais elevadas, menor latência e maior autonomia de bateria nos dispositivos. A MEO seguiu o mesmo caminho e implementou igualmente o 5G SA nas suas antenas de 3,5 GHz.
A menor dimensão da rede 5G da MEO tem levado a uma aposta na eficiência no uso do espectro, em vez do número de sites, ao mesmo tempo em que acelera a expansão (aumento de 30% no número de sites 5G no último ano, o crescimento mais rápido entre os operadores). A estratégia equilibrada assenta em capacidade em banda média e cobertura em banda baixa.
A MEO registou a maior utilização da faixa dos 3,5 GHz (n78) em todas as geografias testadas, operando um canal contínuo de 90 MHz a nível nacional. Este bloco de banda média é o mais amplamente implementado em Portugal: mais de dois terços das amostras da rota nacional recorreram a ele, 76–80% nas ilhas e 88% em Lisboa.
Fora dos grandes centros urbanos como Lisboa e Porto, a MEO faz uso extensivo da banda 700 MHz (n28) como camada de cobertura. A nível nacional, cerca de um terço das amostras 5G foram captadas em 5 MHz dos 700 MHz, enquanto em Lisboa a percentagem caiu para menos de 10% e nas ilhas situou-se entre 20% e 22%. Esta estratégia maximiza o tempo de utilização da banda média e mantém alta a modulação para o utilizador médio, preservando simultaneamente a probabilidade de cobertura nas extremidades das células, o que reduz falhas de acesso e interrupções em vídeo.
Por dispor de um canal de banda baixa mais estreito que o dos rivais (a Vodafone e a NOS utilizam 10 MHz), e de uma pegada 5G mais pequena e menos diversidade espectral além dos 700 MHz e 3,5 GHz, com pouca agregação de portadoras e uso limitado dos 2,1 GHz, os dispositivos na rede da MEO passam menos tempo em 5G do que nas redes da Vodafone e da NOS.
MEO Leads on Call Setup Times, while Vodafone Features Lowest Call Drop Rate
RootMetrics® | 1H 2025 – National Route
Apesar disso, a MEO partilhou a liderança de fiabilidade na rota nacional, apresentando os melhores resultados de acesso e sucesso de tarefas e o tempo de estabelecimento de chamadas mais rápido, cerca de 1,8 segundos. Na Madeira, verificou-se o mesmo padrão: liderança em acesso e sucesso de tarefas, melhor tempo de chamada e fiabilidade global superior. Em Lisboa e no Porto, a MEO esteve lado a lado com a Vodafone em acesso e sucesso de serviços ficou ligeiramente à frente na taxa de quedas de chamadas na capital. A conclusão: uma maior percentagem de tempo em 5G — como acontece com a Vodafone e a NOS — e maior uso de agregação de portadoras não se traduzem automaticamente numa experiência de rede mais consistente.
Uso intensivo de agregação de portadoras amplia a diversidade espectral na rede da Vodafone
No quarto trimestre de 2024, a ampla rede 5G da Vodafone colocou-a ao nível da NOS em termos de tempo em 5G. Nos testes, 84% das amostras da Vodafone basearam-se em 5G a nível nacional, face a 66% da MEO e 27% da DIGI. O perfil da rede da Vodafone espelha o da NOS, mesmo em áreas urbanas partilhadas passivamente, com ambas a demonstrarem um uso intensivo de agregação de portadoras (CA). A Vodafone, ao contrário da NOS, ainda não possui rede 5G SA.
A operadora combina as suas faixas contíguas de 700 MHz (10 MHz) e 3,5 GHz (90 MHz) através de agregação dupla (2CC), criando uma camada agregada de 100 MHz de capacidade. A largura dos 3,5 GHz é idêntica à da MEO e 10 MHz inferior à da NOS. Na rota nacional, a agregação 2CC foi observada em pouco mais de 20% das amostras 5G, subindo para dois terços das amostras em Lisboa e Porto. Nas ilhas, onde a rede é mais dispersa, os testes mostram uma divisão equilibrada entre 700 MHz e 3,5 GHz, com uso limitado de CA.
Different 5G Spectrum Playbooks: Meo Prioritizes 3.5 GHz, NOS and Vodafone Leverage Carrier Aggregation
RootMetrics® | 1H 2025 (Observed Spectrum Use and Bandwidth, National Route)
O uso avançado de agregação pela Vodafone eleva o desempenho tanto de pico como mediano, revelando o melhor desempenho no percentil 5 nacional. Isto reflete partilha eficiente de recursos e forte adaptação de ligação, com seleção otimizada de modulação e codificação, permitindo manter o desempenho nas extremidades das células e em áreas de maior tráfego, mesmo com agregação limitada.
Estes resultados traduzem-se em elevada fiabilidade. A Vodafone co-liderou a fiabilidade global da rede com a MEO e liderou isoladamente nos Açores, registando a menor taxa de chamadas caídas na rota nacional. Em Lisboa e Porto, apresentou a melhor latência de acesso e liderou ou partilhou a liderança em fiabilidade de vídeo, demonstrando que otimizações além da adoção do núcleo SA, como rotas curtas no core e proximidade de CDNs, podem gerar excelentes resultados.
A vantagem de pioneirismo da NOS no 5G SA reforça a capacidade de resposta da rede
Os testes revelaram que a NOS dispõe da configuração de rede mais avançada em Portugal, combinando uma extensa rede 5G SA com a maior variedade de combinações de agregação de portadoras. Opera igualmente a maior rede de sites 5G, com cerca de 4.800 locais no final de 2024, o que a colocou ao nível da Vodafone em tempo passado em 5G (84% a nível nacional).
Tal como a Vodafone, a NOS combina uma largura de banda média de 100 MHz (3,5 GHz) com 10 MHz em 700 MHz, aplicando agregação agressiva tanto em NSA como em SA. Em Lisboa, a agregação 2CC foi observada em 59% das amostras, oferecendo a maior capacidade agregada do país (110 MHz), graças ao bloco de banda média mais amplo. Na rota nacional, 2CC surgiu em 20% das amostras. Ainda assim, a NOS registou uma utilização mais baixa da banda de 3,5 GHz na rota nacional (32% das amostras) face à Vodafone (37%) e à MEO (67,6%).
A vantagem de pioneirismo da NOS no 5G SA está a traduzir-se em maior capacidade de resposta e em início de vídeo mais rápido nas cidades. Em Lisboa, 94% das amostras estavam em 5G e 56% em SA, com início de vídeo em cerca de 530 ms; no Porto, voltou a liderar, com 540 ms. Fora das zonas densas, onde o SA representa apenas 9% do uso de 5G na rota nacional, a NOS manteve a liderança nestas métricas, demonstrando roteamento eficiente no núcleo, interligação otimizada com CDNs (tráfego sai do core próximo do utilizador) e agendamento rápido e de baixa latência, garantindo resposta imediata independentemente do tipo de core.
A ausência de banda baixa leva a DIGI a maximizar a construção em banda média
A expansão da rede própria da DIGI tem sido feita com menos recursos. Sem espectro de 700 MHz para suportar uma camada 5G de grande alcance, a operadora apostou fortemente nas bandas médias, o que resulta em recuos mais frequentes para 4G e acesso 5G irregular em interiores e zonas rurais. Nos testes, apenas 27% das amostras na rota nacional estavam em 5G da DIGI (e apenas 16% nos Açores e 21% na Madeira), subindo para 48% em Lisboa, o que evidencia o carácter urbano da sua implementação inicial.
As implementações 5G da DIGI concentram-se em portadoras TDD relativamente estreitas, e não num canal único e largo em 3,5 GHz, refletindo as limitações de um licenciamento descontínuo. As duas portadoras de 3,5 GHz da DIGI estão separadas por 40 MHz: uma atribuída no leilão de 2021 e outra obtida à NOWO no início deste ano, enquanto a Dense Air mantém o espectro intermédio. A agregação de portadoras pode atenuar algumas desvantagens desta configuração dividida, mas subsiste uma lacuna de eficiência e de suporte de dispositivos face a uma portadora única mais larga, menos favorável do ponto de vista de engenharia de rádio.
NOS and Vodafone Boast Advanced Carrier Aggregation Depth in Lisbon
RootMetrics® | 1H 2025 (Observed Network Architecture Share, Lisbon)
Em Lisboa, 98% das amostras 5G da DIGI basearam-se num bloco de 20 MHz em 2,6 GHz (n41). Na rota nacional, a utilização dividiu-se de forma mais equilibrada entre 3,5 GHz (n78) — maioritariamente blocos de 40 MHz, menos de metade da largura dos concorrentes — e 2,6 GHz (n38/n41), com 20 MHz cada. A DIGI utilizou a faixa de 3,5 GHz mais extensivamente no Porto (37%) do que em Lisboa.
A natureza incipiente da rede própria da DIGI se reflete no desempenho inferior em termos de fiabilidade global, com menores taxas de acesso e de sucesso de tarefas em todas as regiões. Embora já demonstre bons resultados em tempos de estabelecimento de chamada (empatando com a MEO), as piores taxas de queda de chamadas indicam que fatores como rede 5G pouco densa e instabilidade nas transições continuam a afetar o desempenho.
Conclusão: Não existe uma solução única na corrida pela fiabilidade de rede em Portugal
A diversidade nas atribuições de espectro (frequências, larguras e contiguidade), nos perfis de clientes (dimensão, localização e procura de tráfego) e nos ciclos de investimento de rede (expansão inicial, substituição de fornecedor ou simples atualização da RAN) implica que os operadores portugueses precisam de estratégias diferenciadas para competir por meio da fiabilidade.
Apesar destas diferenças, todos partilham ênfase na cobertura em banda baixa para alcance, na capacidade em banda média para profundidade, nas interligações robustas no núcleo da rede e no posicionamento estratégico de CDNs.
A liderança da MEO e da Vodafone em fiabilidade global assenta em abordagens distintas: a MEO aposta na implantação agressiva de um bloco largo de 3,5 GHz, enquanto a Vodafone aplica agregação avançada de forma seletiva, o que prova que não existe uma estratégia única para alcançar fiabilidade superior. Da mesma forma, a NOS, com o seu desdobramentoprecoce do 5G SA, está a melhorar o desempenho em métricas sensíveis à latência, e a DIGI já alcança resultados promissores no tempo de estabelecimentode chamada.
Ookla retains ownership of this article including all of the intellectual property rights, data, content graphs and analysis. This article may not be quoted, reproduced, distributed or published for any commercial purpose without prior consent. Members of the press and others using the findings in this article for non-commercial purposes are welcome to publicly share and link to report information with attribution to Ookla.
Luke Kehoe leads Ookla’s research and thought leadership efforts in Europe.
An electronic engineering alumnus of University College Dublin, Luke has extensive experience collaborating with mobile operators, telecoms vendors, and government agencies in research and advisory roles across Europe. He has contributed to internationally recognised thought leadership publications in areas such as 5G, IoT, open RAN, and edge computing, working with prestigious organisations like the Telecom Infra Project and the World Economic Forum.
Market-led fragmentation has left rail passengers with wildly uneven Wi-Fi experiences across different countries.
Europe and Asia’s rail networks, long heralded as a backbone of economic competitiveness, are now judged not only on punctuality and comfort but on the quality of the digital experience onboard. High-quality train Wi-Fi has shifted from nice-to-have to essential rail infrastructure. Commuters expect a home broadband-like experience for streaming, work calls and gaming while crossing the Swiss Alps or skirting Mount Fuji.
Where countries treat train connectivity as rail infrastructure and pair onboard Wi-Fi with rail-specific infrastructure (trackside, LEO satellite or both), everyday outcomes improve measurably for passengers. This study is the first of its kind to use crowdsourced Ookla Speedtest® data to benchmark country-level train Wi-Fi performance across Europe and Asia.
Key Takeaways:
The gap separating Europe’s best and worst is startling. In Q2 2025, Sweden set the pace for train Wi-Fi in Europe with a 64.58 Mbps median download, followed by Switzerland (29.79 Mbps) and Ireland (26.33 Mbps). Laggards like Spain (1.45 Mbps), the UK (1.09 Mbps) and the Netherlands (0.41 Mbps) featured the poorest outcomes, with download speeds as much as 158 times slower than top-performing Sweden.
Legacy Wi-Fi tech drags many rail networks. Across the European markets studied, nearly two in five connections still run on Wi-Fi 4 (a standard dating to 2009), and ~22% use the lower-capacity, more congestion- and interference-prone 2.4 GHz band. The UK still sees over half of all rail connections on Wi-Fi 4, with 38% on 2.4 GHz. In Poland, rail connections remain almost entirely on Wi-Fi 4 and the 2.4 GHz band.
Band and Wi-Fi gen matter, but backhaul is the real bottleneck. Within-country comparisons show substantial uplifts for 5 GHz vs 2.4 GHz (e.g., +328% in Germany) and Wi-Fi 5 vs Wi-Fi 4 (e.g., +241% in Germany). Yet countries that feature a more modern Wi-Fi mix and thus drive greater use of the 5 GHz band, like Spain and Italy, can still underperform on speeds. This demonstrates that backhaul (i.e., the connection between the train’s roof antennas and the public mobile networks), not just cabin Wi-Fi, is the dominant driver of performance.
Asian rail networks feature modern Wi-Fi mix and lower latency but are not always faster. Taiwan posted the lowest latency and the only material Wi-Fi 6 share (~20%), while Japan and South Korea showed virtually no legacy Wi-Fi 4 or 2.4 GHz usage. Across Asia, typical median download speeds (6-8 Mbps) cluster below Europe’s leaders but above its laggards, reflecting different policy approaches (i.e., greater emphasis on cellular than Wi-Fi).
Policy fingerprints are unmistakable and outweigh topographic and demographic factors. When governments and operators treat mobile networks as core rail infrastructure, and invest in dedicated trackside systems, higher-order MIMO with multi-operator bonded train-mounted antennas, and RF-permeable rolling-stock window retrofits, outcomes improve dramatically.
Fragmented Wi-Fi outcomes reflect different policy attitudes across Europe and Asia
Sweden and Switzerland lead the frontier, puncturing the premise that terrain is destiny
Analysis of Speedtest Intelligence® data reveals Europe’s train Wi-Fi experience is split between a performance frontier and a long tail, with a distribution that resembles two radically different market contexts. Sweden led the continent in Q2 2025 with a median download speed of 64.58 Mbps, more than four times Europe’s country-level median (7.59 Mbps) and over 150 times the Netherlands (0.41 Mbps). This lead extended to upload performance, with Sweden delivering uploads (54.95 Mbps) more than twice as fast as the next fastest country.
It was not always this way. From Q1 2022 to Q1 2024, Wi-Fi performance on Sweden’s train networks was flat at ~2 Mbps down and ~0.7–1.9 Mbps up, placing it in the bottom half of European countries. In Q2 2024, however, there was a clear structural break in the trend, with speeds jumping sharply and continuing to rise through Q1 2025. In practical terms, this means Swedish rail users have moved from a constrained Wi-Fi experience (where even video access was marginal) to a level that supports multi-user carriages with HD streaming and smoother video conferencing.
Sweden Delivers the Fastest Wi-Fi on European Trains by a Wide Margin
Speedtest Intelligence® | Q2 2025
Sweden’s strong performance in mobile coverage along rail corridors has emerged despite challenging conditions, such as long, sparse tracks in the northern regions that face severe winter weather. This success stems from a pragmatic, modular policy framework that delivers targeted state aid where market failures are most evident. For instance, in 2022, the Swedish telecoms regulator PTS allocated €2 million to Telia and Net4Mobility for installing passive, operator-neutral infrastructure in select tunnels. Additionally, rail-specific coverage and capacity obligations were integrated into the 2023 spectrum auction for the 900/2100/2600 MHz bands, setting performance targets to boost capacity on mainlines using the 2100 and 2600 MHz bands while adding new sites for 900 MHz coverage.
In 2023, the Swedish government and PTS proposed that the rail infrastructure operator open access to mobile sites, fibre and power along rights-of-way. It also mandated mapping tunnel coverage, which identified 45 tunnels longer than 300 meters still lacking mobile service, along with developing a comprehensive cost plan. The assessment revealed 630 km of track falling below a 10 Mbps threshold (with a 16 dB margin), prompting efforts to address these gaps through the tunnel support initiatives and rail coverage obligations.
While eclipsed by Sweden for the first time in recent quarters and undergoing a decline in competitiveness, Swiss trains continue to be state of the art in terms of onboard connectivity, delivering median download speeds of 29.79 Mbps in Q2 2025 (albeit down significantly from 85.31 Mbps in Q1 2023, likely reflecting architectural changes or additional congestion). Like Sweden, it represents an exemplary engineering feat for a country characterized by extremely difficult terrain, with Swiss rail operator SBB’s network piercing the Alps with steep approaches, tight valleys, long tunnels, high viaducts and avalanche and rockfall zones.
Northern and Central European Rail Networks Perform Strongest on Wi-Fi Upload Speeds Too
Speedtest Intelligence® | Q2 2025
The Swiss model for onboard connectivity differs markedly from most countries. While SBB offers public Wi-Fi on cross-border services (reflecting the data shared here) and at stations, domestic trains rely primarily on zero-rated mobile data via “SBB FreeSurf” rather than universal onboard Wi-Fi. FreeSurf requires a Swiss SIM and the SBB FreeSurf app; once on board, Bluetooth Low Energy (BLE) beacons in the carriage recognize the device and flag the journey segment, allowing traffic to flow over the public mobile networks without debiting the passenger’s data allowance. SBB then settles the associated data usage with participating mobile operators, effectively subsidizing onboard connectivity.
This model sidesteps the shared onboard Wi-Fi bottleneck and the operating expense of repeaters and cellular backhaul, allowing rail and mobile operators to channel capital into a high-quality radio layer along rail corridors. Its critical limitation is access, however, as onboard connectivity effectively extends only to devices and users with a Swiss-issued SIM, constraining tourists and many business travelers.
Beyond Sweden and Switzerland, other countries that performed well above the European average for download speeds last quarter included Ireland (26.33 Mbps), Czechia (23.36 Mbps) and France (19.12 Mbps). Ireland also recorded the lowest latency of any European country in the period at 40 ms. That strong outcome, despite a disproportionately rural geography, is likely aided by legacy diesel rolling stock. With virtually no electrification and trains operating at lower speeds than many networks on the continent, cellular handovers occur less frequently, which can make better RF outcomes easier to achieve.
Outside Central and Northern Europe, train Wi-Fi slows to a crawl
The performance delta between leading countries and laggards like Spain, the Netherlands and the UK was stark in Q2 2025 and has continued to widen over time. Median download speeds in these countries were as much as 158 times slower than in Sweden in Q2 2025, meaning the average rail passenger connected to a Wi-Fi network in these countries suffers a very poor quality of experience in basic applications like video streaming.
Train Wi-Fi Remains Stuck Firmly in the Slow Lane Across Most European Countries
Speedtest Intelligence® | Q1 2023 – Q2 2025
The UK’s underperformance is not a single-cause issue but the result of weaknesses across multiple layers. At the cabin level, over half of connections still run on Wi-Fi 4, and 38% of samples used the 2.4 GHz band in Q2 2025. This continued reliance on legacy Wi-Fi and the interference-prone, capacity-limited 2.4 GHz band constrains performance regardless of cellular backhaul quality.
Compared with several European peers that organize rail under a single state holding or a clearly empowered state infrastructure manager, the UK has historically split responsibility for stations, services and rolling stock across multiple entities, which complicates collaboration with mobile operators. This friction is easing as GBR reforms bring passenger operations under public control and simplify coordination with state-owned Network Rail. Even so, performance remains weak, reflecting the UK mobile market’s lagging position in network quality (57th globally in the latest Speedtest Global Index™) and the reliance on patchy, incidental public mobile coverage for cellular backhaul.
The Netherlands’ poor train Wi-Fi performance is striking given it ranks in the global top 15 for mobile network quality over the same period, with favorable terrain and high urbanization that enables low-cost coverage along rail corridors. The gap reflects under-investment in the onboard Wi-Fi layer: virtually all connections still use Wi-Fi 4, and usage is very low and has collapsed as passengers shift to their own 5G connections. Dutch rail operator NS has reportedly floated ending the Wi-Fi service if the ministry waives the concession requirement.
Cellular takes precedence over Wi-Fi onboard leading Asian rail networks
Policy muscle in South Korea, Japan and Taiwan has prioritized dedicated trackside cellular coverage, with public Wi-Fi treated more as an amenity than a core service and most passengers relying on their own 4G/5G connections onboard (as in the Netherlands and Switzerland). Even so, rail operators still provide Wi-Fi across much of their rolling stock, and deployments are generally more modern than in Europe.
Wi-Fi 5 and the 5 GHz band are widespread in Japan and South Korea (>90% sample share) on rail networks, with little of the legacy burden seen in countries like the UK or Poland, and Taiwan already features a meaningful and growing share of Wi-Fi 6 (about 20% in Q2 2025) despite still featuring some Wi-Fi 4 (30% sample share).
Taiwan Leads on Latency on the Tracks, Providing a Superior Experience in Interactive Applications
Speedtest Intelligence® | Q2 2025
While none of the studied Asian countries competed at the level of the best European performers in terms of speeds on train Wi-Fi in Q2 2025, each performed well above the long tail of laggards in Europe and close to the average. Taiwan led the pack with median download speeds of 8.1 Mbps in Q2 2025, followed by South Korea (7.11 Mbps) and Japan (6.89 Mbps). The same ranking pattern was observed for upload speeds.
Taiwan delivered the lowest latency of any country in the same period (13 ms), with median response significantly below South Korea (62 ms) and Japan (83 ms).
Rail networks pose one of the most daunting engineering challenges for high-quality Wi-Fi
Rail operators view onboard connectivity as a lever for revenue, loyalty and operations, while policymakers increasingly frame it as part of the digital backbone of national transport systems. The engineering reality is harsher: a train carriage is a metal Faraday cage moving through tunnels, cuttings and rural not-spots, where cellular handovers are frequent and fragile. Best-effort aggregation of public 4G and 5G networks rarely delivers the capacity, stability and latency modern use cases demand.
Delivering a home broadband-like experience on the tracks requires tight coordination across multiple infrastructure layers managed by different entities, typically split into train-to-ground backhaul (via cellular and/or satellite) and on-train distribution systems (via Wi-Fi).
Backhaul still mostly relies on incidental mobile network coverage
The prevailing approach, still used in the vast majority of European countries, relies on wireless backhaul that piggybacks on “incidental” public mobile coverage, feeding dedicated external antennas on each carriage. Because this coverage is incidental, the mobile site grid is usually optimized for nearby population centers rather than the rail corridor itself, creating frequent not-spots and forcing fallback to lower-frequency spectrum with less bandwidth and capacity at cell edges.
Modern Wi-Fi Equipment But Poor Speeds in Countries like Taiwan Indicates Backhaul Problems
Speedtest Intelligence® | Q1 2023 – Q2 2025
On the train itself, regardless of the backhaul feeding the roof-mounted antennas, multi-SIM gateways bond signals from public mobile networks (and, increasingly, LEO providers such as Starlink) and feed an Ethernet backbone to multiple Wi-Fi access points per carriage. Greater bonding diversity across public mobile networks (i.e., using operators with independent infrastructure, not actively shared RAN) typically improves outcomes, since connections can switch dynamically as signal conditions vary. That diversity also adds cost, meaning some rail operators choose a single-network arrangement to contain spend at the expense of performance.
The train carriage itself has become a signal attenuator
The use of external antennas for backhaul is specifically intended to mitigate the fact that rail carriages themselves have become a significant signal attenuator and Faraday cage (and means onboard Wi-Fi can play a complementary role in mitigating against signal loss suffered by 4G and 5G signals on user devices). Modern rolling stock often uses low-E glass with metalized coatings (inducing more signal loss than a layer of concrete in many cases) and foil-backed insulation to reduce heat loss and act as an acoustic barrier. The impact of these RF-hostile designs is compounded at speed, when frequent cell handovers, the Doppler effect, cuttings and tunnels can create jitter (variance in latency over time) and signal dropouts.
Inside the train, crowding adds “body loss” and concentrates hundreds of users onto whatever backhaul is available. This also strains the onboard Wi-Fi, a shared medium whose performance depends on access point placement, channel planning, per-car Ethernet backhaul, and QoS or fair-use policies that may aggressively shape traffic and artificially depress performance.
Leading countries are mobilizing a diverse policy toolkit to deliver better outcomes
Dedicated trackside deployments are needed to tackle cellular not-spots
While cost-effective, leading countries are moving away from the incidental coverage model and converging on dedicated trackside deployments, fostering tighter collaboration between mobile and rail operators to deliver better outcomes. Purpose-built radios along the rail right-of-way, with close inter-site spacing and engineered tunnel coverage using leaky feeders and small cells, allow capacity to scale with corridor demand rather than the surrounding macro grid.
In France, for example, a dedicated trackside layer was introduced on flagship corridors beginning with Paris/Lyon. Orange won an SNCF-run tender to build the network (known as NET.SNCF). Site spacing of ~2–3 km was initially targeted, including the implementation of antenna downtilt and clutter management in cuttings and tunnels, to cater to a TGV (French high-speed train) traveling at 300 km/h handing over base stations as frequent as every 15 seconds.
Notwithstanding the poor performance observed in this study, Austria has employed a similar state-orchestrated, co-funded program since 2015. It has deployed hundreds of mobile sites across 1,500 km of track, initially targeting trackside 4G sites roughly every 5 km and DAS/leaky-feeder systems in tunnels, delivered through a mixture of new-build sites and co-location on existing rail operator ÖBB assets such as GSM-R masts and catenary masts (used to support the overhead electric wires).
Adoption of Higher Wi-Fi Bands Like 5 GHz and 6 GHz Can Improve Performance in Crowded Trains
Speedtest Intelligence® | Q2 2025
Austria’s interventions are based on three-way governance, with ÖBB as the corridor owner and project integrator, mobile operators funding and operating the networks, and the Ministry co-funding and setting expectations via the Rahmenplan (the federal financing instrument that underwrites rail infrastructure programmes in Austria).
In Asia, meanwhile, the Japanese government has subsidized cellular extensions into tunnel segments through a “Radio Shadow Countermeasure Program” with dedicated DAS/relay installations. This means all Shinkansen tunnels have been covered with mobile coverage across NTT Docomo, KDDI and SoftBank since 2020.
Rolling stock retrofits focus on making modern glass less like a layer of concrete
Maximizing returns on dedicated trackside investment means treating the rolling stock as part of the policy toolkit too. Upgrades to the external train-to-ground path focus on multi-band 4×4 (and higher) MIMO and adopting active rooftop antennas powered over Ethernet (PoE). By moving filters and radio components into the antenna radome, operators can avoid long RF coax runs and cut signal losses. Germany’s Deutsche Bahn, for example, used its “advanced TrainLab” program to test and compare rooftop antenna carriers and component combinations, and has since signed a turnkey retrofit and new-build contract with HUBER+SUHNER and McLaren Applied for active PoE rooftop antennas as part of its fleet modernization.
To cut reliance on on-board repeaters and reduce signal attenuation in cellular-based systems (e.g., Switzerland’s SBB FreeSurf) where Wi-Fi is not used, operators have turned to window-replacement programs using laser-treated, RF-permeable low-E glass. Research by EPFL, Swisscom and SUPSI found such windows to be “as good as ordinary glass” for mobile signal, mitigating the 20–30 dB losses recorded by the UK Department for Transport in testing.
Over the last two years, Germany’s Deutsche Bahn announced the laser treatment of 70,000 windows across 3,300 ICE/IC cars (at a cost of €50 million, US$58.7M) and began regional retrofits, following the 2020 decision to equip new high-speed ICE rolling stock with RF-permeable glass as standard. Belgium has pursued a similar policy, abandoning a national on-train Wi-Fi rollout (projected to cost €173 million (US$203M) upfront and €13 million (US $15.3M) in annual operating costs) and redirecting €40 million (US$47M) to alter window coatings and prompt passengers to rely only on their cellular subscriptions while on board.
LEO satellite is emerging as a complement to cellular backhaul for trains
The appeal of low Earth orbit (LEO) for rail operators is increasingly clear. It can add coverage resilience when bonded with cellular on rural, coastal and non-electrified corridors where dedicated trackside and macro layers are thin. LEO’s markedly lower latency and strong burst capacity relative to legacy GEO systems used by many rail operators enables step-change improvements in the onboard passenger Wi-Fi experience and supports operational uses such as CCTV backhaul.
Notwithstanding the opportunity, the constraints of LEO solutions in a rail context are just as real. Hardware maturity still lags aviation and maritime, with far fewer rail-certified, low-profile roof-mount terminals that combine ingress protection, shock and vibration resilience and compliance with EN rail standards, which limits scale for now. Other barriers include sky-view limitations in tunnels and deep cuttings, the operating cost of LEO backhaul for high-demand Wi-Fi unless traffic is shaped and offloaded to cellular, and roof space, power and EMC (electromagnetic compatibility) trade-offs on legacy rolling stock.
Recent commercial and policy developments point to a hybrid end state for LEO on trains, rather than a full replacement for cellular backhaul. Momentum is building in Europe through targeted route trials, limited fit-outs and active procurements, with noticeably less activity in Asia so far. SpaceX’s Starlink and Eutelsat’s OneWeb are the primary LEO constellations in the rail segment, both now in live trials with integrators such as Icomera and CGI, following successful deployments across other transport modes like aviation.
ScotRail, backed by the Scottish Government, has been an early mover with a six-month Starlink pilot on rural northern routes, targeting enhanced passenger Wi-Fi, GPS tracking and live CCTV. In France, SNCF has launched a national tender to equip the fleet with hybrid satellite and terrestrial cellular backhaul, with Eutelsat OneWeb signalling its intent to bid. Italy has ministry-sponsored LEO trials on the Rome to Milan corridor with Trenitalia. PKP Intercity in Poland, České dráhy in Czechia and LTG Link in Lithuania have also tested Starlink terminals to lift onboard Wi-Fi performance.
Policy is converging on using LEO as an additive layer within a multi-link software-defined wide area network ( SD-WAN) gateway onboard that also bonds multiple independent terrestrial cellular networks. In the near term, rail operators will prioritise the corridors with the highest return on investment, need to engineer antenna diversity onboard (for example, two spaced flat-panel terminals to improve link availability through slews, curves and partial obstructions) and issue RFPs that preserve multi-orbit and multi-provider choice with rail-grade certifications for LEO terminals.
Rail connectivity is undergoing a renaissance as satellite and dedicated 5G networks for rail converge
Alongside investments in LEO solutions, rail operators in developed markets are preparing to migrate from legacy GSM-R to Future Railway Mobile Communications System (FRMCS), a 5G-based railway communications standard defined by 3GPP for mission-critical rail. The shift is capital intensive but delivers a dedicated, private 5G trackside network for safety-critical functions such as driver-to-signaler voice, ETCS train control data, remote monitoring and control of trackside assets and live operational and security video.
In Europe, deployments are planned (into the 2030s) primarily in the 900 MHz band with an additional 1.9 GHz capacity layer, and the system will incorporate mission-critical push-to-talk, strict quality of service and, in time, network slicing. While FRMCS focuses on operational communications rather than passenger Wi-Fi or public cellular, the trackside densification it drives is likely to lift the baseline for onboard Wi-Fi by delivering a stronger, more contiguous cellular backhaul layer for bonding.
Together with more capable roof-mounted antennas, RF-permeable window retrofits and Wi-Fi 6E/7 upgrades, these interventions give lagging countries a clear set of levers to lift passenger Wi-Fi performance on board over the coming years.
Ookla retains ownership of this article including all of the intellectual property rights, data, content graphs and analysis. This article may not be quoted, reproduced, distributed or published for any commercial purpose without prior consent. Members of the press and others using the findings in this article for non-commercial purposes are welcome to publicly share and link to report information with attribution to Ookla.
Luke Kehoe leads Ookla’s research and thought leadership efforts in Europe.
An electronic engineering alumnus of University College Dublin, Luke has extensive experience collaborating with mobile operators, telecoms vendors, and government agencies in research and advisory roles across Europe. He has contributed to internationally recognised thought leadership publications in areas such as 5G, IoT, open RAN, and edge computing, working with prestigious organisations like the Telecom Infra Project and the World Economic Forum.
Timely spectrum allocation and proactive policies, not the tyranny of geography or demographics, define Europe’s 5G coverage leaders
Europe is now midway through the 5G technology cycle. Capital spending on network expansion has peaked for most countries, and the flagship low- and mid-band spectrum auctions necessary for 5G deployment are complete. Mobile data traffic growth is now slowing for the first time, and European operators have been more cautious than peers in North America or Asia in adopting new technologies like 5G Standalone (SA), largely due to challenging operating conditions related to sluggish average revenue per user (ARPU) growth.
From a policy perspective, the European Commission has placed 5G at the core of its competitiveness strategy, closely linking coverage availability, timely spectrum assignment, and vendor diversity to productivity gains and strategic autonomy. The EU’s 5G policy agenda is converging on three key imperatives: streamlining infrastructure deployment through initiatives like the Gigabit Infrastructure Act (GIA) and upcoming Digital Networks Act (DNA); subsidizing frontier R&D via programs such as CEF Digital and SNS-JU; and de-risking vendor supply chains through the Security Toolbox and support for open RAN.
This research, which leverages Speedtest Intelligence® data, aims to independently benchmark progress toward the EU’s flagship 5G deployment objectives, including the Digital Decade 2030 goal of achieving 100% outdoor 5G population coverage, using the world’s largest consumer-initiated dataset. It represents the first installment in a three-part series examining Europe’s progress in 5G coverage, network performance, and legacy network sunsets.
Key Takeaways:
Europe’s 5G rollout has produced a “two-speed” competitiveness landscape, with some countries surging ahead in deployment while others fall behind. In Q2 2025, Nordic and Southern European countries maintained a substantial lead in 5G Availability, fueled by recent 700 MHz band deployments that drove double-digit coverage gains in countries such as Sweden and Italy. By contrast, 5G Availability in Central and Western European laggards such as Belgium, the United Kingdom, and Hungary remains less than half that of the leaders. On average, EU mobile subscribers spent 44.5% of their time connected to 5G networks in Q2 2025, up from 32.8% a year earlier.
The deployment and adoption of 5G SA in Europe remain sluggish, increasing slowly from a very low base and further widening the region’s gap with North America and Asia. Spain stands out as a clear leader in 5G SA deployment, reaching an 8% Speedtest® sample share compared with the EU average of just 1.3% as of Q2 2025. This progress has been driven by Spain’s proactive use of EU recovery funds to subsidize 5G SA rollouts in underserved areas, with a particular focus on bridging the rural-urban digital divide. However, the U.S. and China are still far ahead, with 5G SA sample shares above 20% and 80% respectively, reflecting a much greater pace of coverage and adoption in those markets.
Fragmented 5G Availability across Europe is driven by a complex mix of national policies on spectrum assignment and broader economic factors, rather than by simple geographic or demographic differences. 5G Availability is more strongly correlated with policy-driven factors such as spectrum allocation timelines and costs, coverage obligations, subsidy mechanisms, and regulations for infrastructure sharing and permitting, than with structural factors like urbanization rates or the number of operators. This indicates that 5G competitiveness is shaped less by technology gaps or inherent market imbalances and more by effective policy execution.
Europe’s recent fulfillment of the 5G pioneer band strategy masks fragmentation
This year is the first time that the EU’s “pioneer bands” for 5G, identified in the Commission’s 5G Action Plan to support early harmonized spectrum availability, have been substantially assigned across the bloc. With recent auctions in Poland for low-band and the Netherlands for mid-band, every member state except Malta has now allocated 60 MHz in the 700 MHz band and 400 MHz in the 3.4-3.8 GHz band for 5G. This effectively completes the 5G auction pipeline in Europe until demand increases for the final pioneer band, the 26 GHz mmWave band (1,000 MHz), which is likely to be used primarily for capacity in-fill in very dense urban environments like stadiums.
This important milestone in assignment harmonization marks the end of nearly a decade of significant fragmentation in spectrum availability for 5G across Europe, which had undermined the conditions needed for the Commission’s pursuit of a single market for telecom. For example, there was almost a nine-year gap between the 700 MHz assignment in Finland, one of the first movers in 2016, and in Poland, which only completed its assignment earlier this year, despite both countries having committed to the same Digital Decade targets.
Northern Europe Maintains 5G Availability Lead, Benelux and Eastern Europe Lag
Speedtest Intelligence® | Q2 2025
Fragmentation remains a persistent theme, shaping stark 5G deployment asymmetries that cannot be explained by geography or demographics alone. Northern and Southern European countries such as Denmark (83.9%), Sweden (77.8%), and Greece (76.4%) are disproportionately represented among the countries with the highest 5G Availability in Q2 2025, with coverage rates up to twice as high as those in Western and Eastern countries like the United Kingdom (45.2%), Hungary (29.9%), and Belgium (11.9%).
Northern and Southern Europe lead in 5G Availability through a carrot-and-stick mix of spectrum management, subsidies, and coverage obligations
Nordic countries such as Denmark, Sweden, and Norway—with Sweden and Norway featuring some of the lowest population densities and most challenging terrain in Europe—continue to distinguish themselves in the 5G cycle through innovative policy approaches. All three have imposed stringent rural or regional coverage obligations on 5G spectrum licenses. For example, in Sweden’s 700 MHz band auction, Telia was required to invest €25 million (US$29 million) from its license fee to provide at least 10 Mbps coverage in prioritized rural areas lacking adequate service, with operators aiming for 99% nationwide population access by the end of this year.
These Nordic countries have also actively promoted extensive network sharing, such as the TT Network joint venture between Telia and Telenor in Denmark and Net4Mobility between Tele2 and Telenor in Sweden, and leveraged loans from the European Investment Bank (EIB) or Nordic Investment Bank (NIB) to fund rural rollouts and support early 700 MHz deployments to create a “true” 5G coverage layer rather than relying solely on dynamic spectrum sharing (DSS).
Similarly, Switzerland continues to outpace its Central European neighbors, such as Luxembourg and Belgium, in 5G Availability, reaching 81.3% in Q2 2025. This achievement was realized without government subsidies, relying instead on early, competitively priced access to the pioneer bands and voluntary commitments from operators like Swisscom to deliver extensive 5G coverage (e.g., 90% population coverage by 2024). Affordable spectrum allocation preserved operators’ capital for network investments, bolstered by exceptionally high average revenue per user (ARPU) levels.
Policy acts as a barrier, not a catalyst, for 5G deployment in Western and Eastern European laggards
While regulatory policies have spurred 5G investment in Northern and Southern European countries, they have stifled it in others. In the United Kingdom, the enforcement of the Telecoms Security Act has compelled operators to undertake an expensive rip-and-replace program for vendor equipment in 5G networks by 2027, driven by concerns over supply chain vulnerabilities (with similar impact observed in Hungary). Additionally, the country’s 700 MHz and 3.6-3.8 GHz spectrum auction in 2021 omitted stringent coverage obligations after operators agreed to the voluntary Shared Rural Network (SRN) initiative, which emphasized improving rural 4G coverage rather than accelerating 5G rollout.
These challenges have been compounded by post-Brexit funding gaps, which have prevented the United Kingdom from accessing EU Recovery and Resilience Facility resources, including the €2 billion (US$2.3 billion) allocated for 5G deployment in Italy and the support provided for Spain’s Digital Spain Agenda 2025. At the same time, the country’s operators have been under further pressure from ARPU erosion due to fierce price competition in a four-player market (now changing) and from rising operational costs, especially higher energy prices.
The United Kingdom is not unique in its struggles. Belgium, home to the core of European bureaucracy, still features lower 5G Availability than many emerging markets in Latin America and Southeast Asia. The country’s federal structure led to chronic delays, as spectrum auctions originally planned for 2019 were pushed to 2022 amid regional disputes over revenue sharing between Flanders, Wallonia, and Brussels. Strict radiation limits in Brussels further slowed 5G deployment post-auction.
Analysis of the relationship between 5G Availability and spectrum auction timing in Europe confirms that policy can act as a barrier to deployment when it unduly delays spectrum release. Many operators have used techniques such as DSS to accelerate 5G rollouts before dedicated pioneer bands were available (resulting in some artificial overperformance in countries such as Ireland and Poland). However, the evidence clearly shows that countries which assigned pioneer bands earlier have achieved higher levels of 5G Availability today.
Low-band deployment and DSS use continue to lift 5G availability in lagging countries
Recent advances in 5G Availability have been driven by low-band deployments and the use of DSS, raising the average proportion of time spent on 5G networks in the EU from 32.8% in Q2 2024 to 44.5% in Q2 2025. The pace of coverage growth, and the corresponding increase in 5G usage, has primarily reflected each country’s starting point. Lagging countries like Latvia, Poland, and Slovenia have seen double-digit gains in 5G Availability from a low base. By contrast, leading countries such as Switzerland and Denmark, where 5G coverage is now nearly ubiquitous, have shifted their focus to targeted capacity upgrades through site densification and mid-band expansion.
Significant 5G coverage gains in Sweden (+21.3% YoY in 5G Availability) over the past year have been driven by aggressive 700 MHz deployment by Telia and Tele2 to close rural-urban gaps across the country’s expansive forested terrain, with fiscal backing from government digital inclusion subsidies. In Italy (+20.5% YoY), momentum has come from 3G sunsets (with WindTre repurposing the 2100 MHz band recently), mobilization of EU Recovery Funds through the country’s flagship National Recovery and Resilience Plan (PNRR), and April 2024 policy reforms easing EMF restrictions to facilitate faster infrastructure rollout. Meanwhile, Malta’s operators have rapidly expanded 5G coverage through DSS and benefited from the country’s compact geography, despite still lacking a 700 MHz assignment for 5G.
Low-Band Deployment and DSS Fuel 5G Coverage Expansion in Lagging Countries
Speedtest Intelligence® | Q2 2024 – Q2 2025
Discover how spectrum policy and strategy shape 5G coverage across Europe, with Nordic and Southern nations leading and Spain ahead in 5G SA.
While the European Commission has not yet embedded technology-specific deployment goals for 5G SA in its 2030 Digital Decade policy program, it is now distinguishing the technology from the Non-Standalone (NSA) architecture in several key policy documents, funding initiatives, and monitoring reports. This distinction is often framed as an enabling requirement in the context of helping to boost EU competitiveness, closing innovation gaps, and addressing Europe’s lag behind the U.S. and China in advanced connectivity deployment.
However, first-of-its-kind research published by Ookla earlier this year in collaboration with Omdia revealed that the bloc has fallen far behind in 5G SA deployment. Real-world penetration of the technology, shaped by a combination of network coverage, device adoption, and tariff configuration, remains much lower than headline population coverage figures suggest. By Q2 2025, the competitiveness gap had widened further, with 5G SA sample share (a proxy for coverage) reaching just 1.3% in the EU. This is several times lower than the more than 20% observed in the U.S. and 80% in China in the same period.
Spain's Subsidy-Heavy Policy Framework Drives 5G SA Deployment in Underserved Areas
Speedtest Intelligence® | Q2 2024 – Q1 2025
Spain continues to lead Europe in 5G SA deployment, with its 5G SA sample share surpassing 8% for the first time in Q1 2025. Both MasOrange and Telefónica have driven an aggressive nationwide rollout using a diversified spectrum strategy across low- and mid-bands, extending 5G SA coverage deeper into rural and underserved areas than anywhere else in Europe. This progress has been enabled by Spain’s subsidy-heavy policy framework, which has allocated hundreds of millions of euros from EU recovery funds (NextGenerationEU) through “UNICO-5G” grants to finance more than 7,000 new sites in villages and along 30,000 km of roads.
Key European economies such as the United Kingdom and Germany are achieving stronger progress in 5G SA deployment than their overall 5G Availability figures, which are heavily skewed by NSA networks, might indicate. The United Kingdom’s Wireless Infrastructure Strategy sets out a national ambition, rather than prescriptive obligations, to achieve 5G SA coverage in all populated areas by 2030. This target is among the most ambitious of any advanced liberal economy globally. The country has also leveraged remedies addressing competition concerns over the VodafoneThree merger to require the merged entity to extend 5G SA coverage to 99% of the UK population by 2030.
Meanwhile, the German telecom regulator BNetzA has promoted competition in the 5G SA rollout by being one of the first globally to transparently track 5G SA deployment with detailed operator-level coverage maps available to the public.
Evidence-based policymaking is central for Europe’s competitiveness in frontier technologies like 5G
Persistent disparities in 5G coverage and long delays in harmonizing spectrum availability show that upcoming regulatory initiatives like the DNA face a tall order to improve Europe’s competitiveness in 5G deployment. Yet the experience of member states that moved early on strategic spectrum allocation and applied data‑driven policy levers to spur deployment, often overcoming geographic and demographic disadvantages traditionally seen as impediments, demonstrates that Europe already has the tools needed to close the gap.
Coming next in this three-part series: a Europe‑wide 5G performance study spanning QoS (speeds, latency) and QoE (browsing, video, gaming) built on the world’s largest consumer‑initiated dataset. Stay tuned.
Ookla retains ownership of this article including all of the intellectual property rights, data, content graphs and analysis. This article may not be quoted, reproduced, distributed or published for any commercial purpose without prior consent. Members of the press and others using the findings in this article for non-commercial purposes are welcome to publicly share and link to report information with attribution to Ookla.
Luke Kehoe leads Ookla’s research and thought leadership efforts in Europe.
An electronic engineering alumnus of University College Dublin, Luke has extensive experience collaborating with mobile operators, telecoms vendors, and government agencies in research and advisory roles across Europe. He has contributed to internationally recognised thought leadership publications in areas such as 5G, IoT, open RAN, and edge computing, working with prestigious organisations like the Telecom Infra Project and the World Economic Forum.
From Vulnerability to Resilience: How Portugal’s Mobile Networks Handled the Iberian Peninsula Blackout | Da Vulnerabilidade à Resiliência: Como as Redes Móveis Portuguesas Reagiram ao Apagão na Península Ibérica
Robust power redundancy markedly reduced outage impacts for one operator, while limited backup systems led to widespread service collapse for another, highlighting the importance of resilience planning and investment.
Mobile operators, equipment vendors, and policymakers throughout Europe are grappling with the challenge of hardening telecom infrastructure to withstand increasingly frequent and severe disruptions caused by power outages, sabotage, and extreme weather events.
Earlier this year, the Iberian grid blackout placed Portugal’s mobile operators at the coalface of this resilience challenge, creating a real-world stress test of their infrastructure on an unprecedented scale. Effective power redundancy, supported by battery and generator backups, coupled with energy conservation measures that strategically adjusted network configurations to preserve site availability, emerged as critical tools for limiting outage impact.
However, new analysis of Ookla® background signal scan data from the outage reveals that each operator’s ability to mitigate the disruption varied significantly, offering important lessons for future improvements in Portugal and beyond. This research builds upon our earlier findings in Spain, where we cross-referenced crowdsourced ‘no service’ data with satellite imagery to demonstrate that the profile of network disruptions and recovery moved in lockstep with power grid developments.
Key Takeaways:
At the height of the network disruptions on the evening of April 28th, more than one in three mobile network users in Portugal was left without service. The voltage drop triggered by the grid collapse rapidly cascaded through Portugal’s mobile networks, driving the share of users experiencing total service loss (unable to call, text, or use data as sites went dark) from a pre-blackout baseline below 0.1% to over 10% within two hours. At the peak late on April 28th, as battery and generator backups were progressively depleted, more than 60% of users across the worst-affected areas of Portugal were left without service.
While severe network outages affected all Portuguese operators during the blackout, mobile users on DIGI’s network were significantly more likely to experience a total loss of service. With up to 90% of DIGI subscribers left without any mobile coverage for over twenty-four hours, the outage exposed critical gaps in redundancy across multiple infrastructure layers, from mobile sites at the edge all the way to the core, potentially reflecting the limitations of DIGI’s less mature network buildout in Portugal.
MEO’s network demonstrated significantly greater resilience across Portugal during the April 28th blackout, illustrating how deep and widely deployed battery reserves can materially flatten and delay outage impacts triggered by power loss. At the peak of service disruption six to eight hours after the power loss, MEO’s subscribers were on average half as likely to lose service as those on NOS’s network, four times less likely than Vodafone’s subscribers, and six times less likely than DIGI’s. As a result, at least tens of thousands more MEO subscribers likely stayed connected for calls, texts, and data throughout April 28th.
The variation in outage impact between operators in Portugal was significantly greater than in Spain, revealing much deeper asymmetry in the level of power resilience across Portugal’s mobile networks. As in Spain, however, the pattern of service restoration reflected the geographically phased re-energisation of the power grid, with network disruptions persisting later into the night in Lisbon than in Porto, consistent with transmission operator REN’s blackstart process, which began in the north and moved south.
Blackout cascaded through Portugal’s mobile networks, forcing aggressive energy conservation measures as traffic demand surged and power backups were depleted
When the grid-wide collapse severed power to virtually all of mainland Portugal at 11:33 local time on April 28th, mobile sites were immediately forced off mains electricity and had to rely on batteries or generator backups, triggering a nationwide race between grid restoration and the exhaustion of backup reserves across telecom networks. Sites lacking any power autonomy vanished immediately (such as dense urban small cells), triggering a stepwise collapse in overall network density that resembled a cliff drop followed by a gradually declining tail.
The sudden loss of residential electricity rendered fixed networks and in-home Wi-Fi CPEs unusable, forcing users onto mobile networks and unleashing a massive surge in traffic that put intense pressure on capacity, particularly in urban areas. This was reflected in a rapid degradation of mobile network performance across all metrics, as illustrated in analysis of Speedtest Intelligence® data published in our earlier research.
The spike in demand on the country’s mobile infrastructure occurred just as operators were racing to implement aggressive energy conservation measures to extend the life of backup power at mobile sites. These efforts included phased 5G switch-offs (as 3.5 GHz massive MIMO radios typically draw two to three times the power of a low-band 4G sector), prioritizing core voice and text services, and reducing cell-edge transmit power where network loads were light.
Blackout produced a composite outage curve made of one large step (DIGI) superimposed on several peaked pulses (Vodafone, NOS, and MEO)
Although all of Portugal’s mobile operators implemented similar energy conservation measures during the blackout, the depth and distribution of power autonomy within each operator’s site portfolio, including the partially shared footprint between NOS and Vodafone, ultimately shaped their network resilience. This is evident in the distinct outage trajectories revealed by analysis of background signal scan data, which shows whether a device could connect to any network (2G, 3G, 4G, or 5G) based on a very large, geographically diverse sample across Portugal.
DIGI’s still-nascent network, which is leaner and heavily concentrated in cities (therefore making deployment of power autonomy more challenging at space-constrained rooftop sites), proved particularly brittle. Within four hours of the voltage drop, the share of subscribers on its network with no signal shot up from less than 0.1% to more than 90%, a classic step-function collapse. The operator’s entire radio layer appeared to disappear almost simultaneously, driven by shallow site-level batteries and little layered fallback. In addition, network access remained crippled for more than a day, likely pointing to a catastrophic failure of deeper elements such as the Evolved Packet Core (EPC) in Lisbon, which may have lacked geo-redundancy or sufficient power autonomy.
While Vodafone’s outage curve did not exhibit the same cliff-like profile as DIGI’s, instead following more of a triangular or peaked pulse shape, it still reached a very sharp peak. The heterogeneous distribution of backup power across Vodafone’s site footprint produced a multi-step survival curve, with each autonomy band expiring (for example, sites with four-hour batteries) causing another visible kink in the aggregate outage trajectory.
By 19:30 local time, almost 70% of Vodafone’s subscribers were left without service as the last reserves of backup power began to deplete ahead of grid restoration. While this was still materially lower than the more than 90% service loss seen on DIGI’s network in Portugal, it was nearly twice as high as the peak outage experienced by any operator in Spain on April 28th. Service was, however, rapidly restored on Vodafone’s network from 20:00 in a phased geographic sequence, aligning with the restoration of the grid, with the no service ratio falling below 5% by midnight.
Unlike other operators in Portugal, Vodafone and NOS have extensive RAN sharing, with a joint venture owning and operating actively shared sites in rural and interior areas, while sites in urban areas are passively shared. Despite this, the outage profile for NOS was notably less severe. This indicates that NOS’s network features relatively deeper power resilience in locations where its infrastructure is not actively shared, compared with Vodafone’s independently managed sites. On NOS’s network, the proportion of subscribers without service peaked early at nearly 30%, closely resembling the impact profile of the worst-affected operator in Spain, and remained at this level until power was restored.
The merits of widely deployed and deep battery reserves in flattening and delaying the outage curve (much like masks and vaccines suppress infection spread during a pandemic) were clearly demonstrated in MEO’s case. Its outage peak was lower and the tail shorter, with the proportion of subscribers left without service peaking at just over 16%, which was the best performance observed across Spain and Portugal on April 28th.
Outage experience demonstrates the role of power autonomy and geo-redundancy in hardening telecom infrastructure against external shocks
When the grid collapsed, every Portuguese operator reached for the same first lever by killing off the power-hungry 5G layer, but what happened next diverged. The breadth and depth of each operator’s power autonomy (at the site level) and the extent of geo-redundancy (at the core level), along with their ability to cascade lower-band layers, throttle traffic, and reshuffle spectrum, dictated how much of their network stayed online and for how long during the blackout.
The pronounced asymmetry in outage impacts observed across operators’ subscriber bases highlights the urgent need to harden mobile networks and raise all infrastructure layers to a higher baseline of resilience ahead of future severe events. There is now broad consensus, which is expected to be enshrined in the European Commission’s forthcoming Digital Networks Act (DNA), that telecom networks are critical infrastructure essential for societal functioning, and that even brief service disruptions can quickly escalate into serious public safety risks.
Da vulnerabilidade à resiliência: Como as redes móveis portuguesas reagiram ao apagão na Península Ibérica
Uma forte redundância energética atenuou significativamente os efeitos da falha para um dos operadores, enquanto a escassez de sistemas de reserva provocou a interrupção generalizada dos serviços noutro, evidenciando a importância do planeamento e do investimento em resiliência.
Os operadores móveis, fornecedores de equipamentos e reguladores em toda a Europa estão a enfrentar o desafio de reforçar a infraestrutura das telecomunicações para resistir a interrupções cada vez mais frequentes e graves causadas por falhas de energia, sabotagem e fenómenos meteorológicos extremos.
No início deste ano, o apagão da rede ibérica colocou os operadores móveis portugueses na linha da frente deste desafio de resiliência, criando um teste real de resistência das infraestruturas numa escala sem precedentes. A redundância energética eficaz, apoiada por baterias e geradores de reserva, aliada a medidas de poupança de energia que ajustaram estrategicamente as configurações da rede para preservar a disponibilidade dos sites, revelou-se uma ferramenta crucial para limitar o impacto das falhas de energia.
No entanto, uma nova análise dos dados de monitorização passiva de sinal da Ookla® durante a falha revela que a capacidade de cada operador para mitigar a interrupção variou significativamente, oferecendo lições importantes para futuras melhorias em Portugal e além-fronteiras.
Esta investigação baseia-se nas conclusões anteriores obtidas em Espanha, onde cruzámos dados crowdsourced de “sem serviço” com imagens de satélite para demonstrar que o perfil das perturbações e da recuperação das redes evoluiu em paralelo com a situação da rede elétrica.
Principais conclusões:
No auge das perturbações na rede, na noite de 28 de abril, mais de um em cada três utilizadores de redes móveis em Portugal ficou sem serviço. A queda de tensão desencadeada pelo colapso da rede elétrica propagou-se rapidamente nas redes móveis do país, fazendo com que a proporção de utilizadores com perda total de serviço (sem possibilidade de fazer chamadas, enviar mensagens ou utilizar dados, à medida que os sites ficavam inoperacionais) subisse de um valor inferior a 0,1 % antes do apagão para mais de 10 % em menos de duas horas. No pico, já no final do dia de 28 de abril, à medida que as baterias e os geradores de reserva se esgotavam progressivamente, mais de 60 % dos utilizadores nas zonas mais afetadas de Portugal ficaram sem serviço.
Embora todas as operadoras portuguesas tenham sido afetadas por graves falhas na rede durante o apagão, os utilizadores móveis da rede DIGI foram significativamente mais propensos a experienciar uma perda total de serviço. Com até 90 % dos assinantes da DIGI sem qualquer cobertura móvel durante mais de vinte e quatro horas, a falha expôs lacunas críticas na redundância em vários níveis da infraestrutura, desde as antenas móveis na periferia até ao núcleo da rede, refletindo potencialmente as limitações do desenvolvimento menos amadurecido da rede da DIGI em Portugal.
A rede da MEO demonstrou uma resiliência significativamente maior em todo o território português durante o apagão de 28 de abril, mostrando como as reservas robustas e amplamente implantadas de baterias podem atenuar e atrasar de forma significativa os impactos das falhas de energia. No pico da interrupção do serviço, entre seis e oito horas após a perda de energia, os assinantes da MEO tinham, em média, metade da probabilidade de perder o serviço comparativamente aos da NOS, quatro vezes menos do que os da Vodafone e seis vezes menos do que os da DIGI. Como resultado, provavelmente dezenas de milhares de assinantes da MEO mantiveram-se conectados para chamadas, mensagens e dados ao longo de todo o dia 28 de abril.
A variação do impacto das interrupções entre operadores em Portugal foi significativamente maior do que em Espanha, revelando uma assimetria muito mais profunda no nível de resiliência energética das redes móveis portuguesas. No entanto, tal como em Espanha, o padrão de restabelecimento do serviço refletiu a reenergização faseada geograficamente da rede elétrica, com as perturbações a persistirem até mais tarde durante a noite em Lisboa do que no Porto, em conformidade com o processo de arranque da rede de transporte da REN, que começou no norte e avançou para sul.
O apagão propagou-se pelas redes móveis de Portugal, obrigando a medidas agressivas de poupança de energia, à medida que a procura de tráfego aumentava e as reservas de energia se esgotavam
Quando o colapso de toda a rede cortou a energia em praticamente todo o território português continental, às 11h33, hora local, do dia 28 de abril, as antenas móveis foram imediatamente desligadas da corrente elétrica principal e tiveram de recorrer a baterias ou geradores de reserva, desencadeando uma corrida nacional entre a restauração da rede e o esgotamento das reservas de energia nas redes de telecomunicações. As infraestruturas que não dispunham de qualquer autonomia energética desapareceram imediatamente (como as “small cells”— micro- e pico-células), desencadeando um colapso gradual da densidade global da rede que se assemelhou a uma queda abrupta seguida por uma diminuição gradual.
A súbita perda de eletricidade residencial tornou as redes fixas e os equipamentos de Wi-Fi domésticos (CPE) inutilizáveis, obrigandoos utilizadores a recorrer às redes móveis e desencadeando um aumento massivo de tráfego que exerceu uma pressão intensa sobre a capacidade, em especial nas áreas urbanas. Isto refletiu-se numa rápida degradação do desempenho das redes móveis em todas as métricas, conforme ilustrado na análise dos dados Speedtest Intelligence® publicada na nossa investigação anterior.
O aumento súbito da procura na infraestrutura móvel do país ocorreu precisamente quando as operadoras estavam a correr para implementar medidas agressivas de poupança de energia para prolongar a duração da energia de reserva nos sites móveis. Esses esforços incluíram o desligamento faseado do 5G (já que os rádios MIMO de 3,5 GHz normalmente consomem duas a três vezes a potência de um setor 4G de baixa frequência), priorizando os principais serviços de voz e texto e reduzindo a potência de transmissão de ponta das células quando as cargas de rede eram baixas.
O apagão produziu uma curva de falha composta por um grande salto (DIGI) sobreposto a vários picos distintos (Vodafone, NOS e MEO)
Embora todos os operadores móveis em Portugal tenham implementado medidas semelhantes de poupança de energia durante o apagão, o grau e a distribuição da autonomia energética da infraestrutura de rede de cada operador, incluindo a infraestrutura parcialmente partilhada entre a NOS e a Vodafone, acabaram por moldar a resiliência das suas redes. Isso é evidente nas trajetórias distintas de falhas reveladas pela análise dos dados de monitorização passiva de sinal, que indicam se um dispositivo conseguia ligar-se a uma das redes (2G, 3G, 4G ou 5G), com base numa amostra ampla e geograficamente diversificada de todo o território português.
A rede ainda em fase inicial da DIGI, com uma cobertura mais limitada e fortemente concentrada em áreas urbanas (o que torna a implantação da autonomia de energia mais difícil em infraestruturas no topo de edifícios), revelou-se particularmente frágil. Quatro horas após o colapso da rede elétrica, a percentagem de assinantes da sua rede sem qualquer sinal disparou de menos de 0,1 % para mais de 90 %, um colapso abrupto e generalizado, típico de um corte súbito. Toda a infraestrutura de rádio da operadora parece ter desaparecido quase em simultâneo, impulsionada por baterias de curta duração ao nível dos sites e com pouca redundância em camadas superiores. Além disso, o acesso à rede permaneceu severamente comprometido por mais de um dia, apontando provavelmente para uma falha catastrófica de elementos mais profundos, como o Evolved Packet Core (EPC) em Lisboa, que pode ter carecido de redundância geográfica ou de autonomia energética suficiente.
Embora a curva de falhas da Vodafone não tenha apresentado o mesmo perfil de queda abrupto como a DIGI, seguindo mais uma forma triangular, ainda assim atingiu um pico muito acentuado. A distribuição heterogénea da autonomia energética na rede em toda a área de cobertura da Vodafone produziu uma curva de sobrevivência em várias etapas, com cada faixa de autonomia a esgotar-se (por exemplo, locais de rede com baterias de quatro horas) a provocar um novo ressalto visível na trajetória do apagão.
Por volta das 19h30, hora local, quase 70 % dos assinantes da Vodafone estavam sem serviço, à medida que as últimas reservas de energia de apoio começaram a esgotar-se antes do restabelecimento da rede elétrica. Embora este valor seja significativamente inferior aos mais de 90 % de perda de serviço verificados na rede da DIGI em Portugal, representava quase o dobro do pico de interrupção registado por qualquer operador em Espanha no dia 28 de abril. O serviço, contudo, foi rapidamente restabelecido na rede da Vodafone a partir das 20h00, seguindo uma sequência geográfica faseada, em consonância com a reposição da rede elétrica, com a taxa de ausência de serviço a cair para menos de 5 % à meia-noite.
Ao contrário de outros operadores em Portugal, a Vodafone e a NOS partilham extensivamente a RAN, uma joint venture que possui e opera infraestruturas partilhadas ativamente em áreas rurais e do interior, enquanto nas zonas urbanas as infraestruturas são partilhadas de forma passiva. Apesar disso, o perfil de interrupções da NOS foi notavelmente menos grave. Isto indica que a rede da NOS apresenta uma resiliência energética relativamente maior nos locais onde a sua infraestrutura não é ativamente partilhada, em comparação com os geridos de forma independente pela Vodafone. Na rede da NOS, a proporção de subscritores sem serviço atingiu um pico precoce de quase 30 %, assemelhando-se de perto ao perfil de impacto do operador mais afetado em Espanha, mantendo-se neste nível até a energia ser restabelecida.
Os méritos das reservas de baterias amplas e profundamente distribuídas na atenuação e no adiamento da curva de falhas (de forma semelhante às máscaras e vacinas na contenção da propagação de infeções durante uma pandemia) ficaram claramente demonstrados no caso da MEO. O pico da interrupção foi menor e a retoma mais rápida, com a proporção de assinantes sem serviço a atingir um pouco mais de 16 %, o melhor desempenho observado em toda a Península Ibérica no dia 28 de abril.
A experiência do apagão demonstra o papel da autonomia energética e da geo-redundância no reforço da resiliência da infraestrutura das telecomunicações face a choques externos
Quando a rede elétrica colapsou, todos os operadores portugueses tentaram a mesma primeira alavanca: desligar a camada de 5G, intensiva em consumo energético. Mas a partir daí, os caminhos divergiram. A extensão e a robustez da autonomia energética de cada operador (ao nível das infraestruturas) e o grau de geo-redundância (ao nível do núcleo), juntamente com a capacidade de cascatear camadas em bandas inferiores, limitar o tráfego e reconfigurar o espectro, ditaram quanto da sua rede permaneceu operacional, e durante quanto tempo, durante o apagão.
A acentuada assimetria nos impactos do apagão observada entre as bases de clientes dos diferentes operadores realça a necessidade urgente de reforçar as redes móveis e elevar todos os níveis da infraestrutura a um patamar mais robusto de resiliência, em antecipação a futuros eventos graves. Existe um consenso alargado, que se prevê que venha a ser consagrado no futuro Digital Networks Act (DNA) da Comissão Europeia, de que as redes de telecomunicações são infraestruturas críticas essenciais para o funcionamento da sociedade, e que mesmo interrupções breves do serviço podem rapidamente transformar-se em riscos sérios para a segurança pública.
Ookla retains ownership of this article including all of the intellectual property rights, data, content graphs and analysis. This article may not be quoted, reproduced, distributed or published for any commercial purpose without prior consent. Members of the press and others using the findings in this article for non-commercial purposes are welcome to publicly share and link to report information with attribution to Ookla.
Luke Kehoe leads Ookla’s research and thought leadership efforts in Europe.
An electronic engineering alumnus of University College Dublin, Luke has extensive experience collaborating with mobile operators, telecoms vendors, and government agencies in research and advisory roles across Europe. He has contributed to internationally recognised thought leadership publications in areas such as 5G, IoT, open RAN, and edge computing, working with prestigious organisations like the Telecom Infra Project and the World Economic Forum.
The UK’s biggest music festival presents a unique but challenging platform for mobile operators to showcase the strength of their networks
This year’s Glastonbury Festival once again transformed Worthy Farm, a quiet corner of rural England, into the UK’s seventh-largest city for a long weekend of music and celebration in blistering heat. With over 200,000 festival-goers descending on the 900-acre site, all eager to stay connected, the event becomes an immense stress test for the UK’s mobile operators—requiring them to build a temporary metropolitan-scale network from scratch in under three weeks, only to dismantle it again on Monday morning.
In many ways, Glastonbury is the Super Bowl for operators. It offers a once-a-year opportunity to showcase their network pedigree and technical prowess under the spotlight of intense public scrutiny. In this article, we examine how the networks performed during the big event using Speedtest® and background signal scan data collected throughout Worthy Farm.
Key Takeaways:
Three’s spectrum advantage translated into a significant lead in network performance across key metrics during Glastonbury 2025. Median mobile download speeds on Three (347.66 Mbps) were at least twice as fast as those on any other operator, and the network also topped quality of experience (QoE) measures reflecting the typical performance for web browsing, video calling, and gaming throughout the event. Three’s structural advantage, particularly its larger contiguous mid-band allocation (roughly twice the width of its rivals) and leaner subscriber base (with fewer users per antenna sector), was likely central to its performance lead.
Festival-goers were least likely to experience poor performance on EE’s network throughout Worthy Farm. At the 10th percentile, which captures the slowest 10% of outcomes when signal was weakest or congestion highest, EE still recorded the fastest download and upload speeds of any operator. For festival-goers, this meant fewer moments of buffering; even during peak crowd surges they were more likely to be able to keep streaming and uploading on EE’s network. This likely reflects the performance benefits conferred by the operator’s unique spectrum diversity, which has enabled it to leverage a broad carrier aggregation mix by combining multiple low and mid-band carriers.
Vodafone and O2 lagged on key metrics, though both delivered clear improvements in network performance compared to previous festivals. O2’s lagging position is likely explained by its combination of the largest subscriber base (meaning more users per antenna sector) and the smallest deployable spectrum portfolio, particularly for mid-band 4G. This contributed to O2 trailing its rivals in median download speed performance at the festival. Vodafone fared better on most metrics but recorded the weakest download speed outcomes at the 10th percentile, suggesting more acute congestion issues on its network. Even so, both operators saw median download speeds rise by at least 25% year-on-year, indicating that network investments are paying off.
Glastonbury has become a seminal branding and engineering showcase for operators looking to tap its vast social reach
Music festivals like Glastonbury skew young, digitally driven, and high in disposable income, exactly the demographic operators fight hardest for. As a result, the event has become a strategic trifecta for operators, acting as an engineering showcase, a customer experience litmus test, and a flagship moment in a summer-long calendar of music-led brand activations all at once.
From a branding perspective, both Vodafone (currently) and EE (previously) have sought to capitalize on this by investing heavily in event sponsorship to secure the “Official Connectivity Partner” title. This status brings exclusivity on-site and within the official app, prime logo placement on stage screens and TV broadcasts, and control over experiential zones such as Vodafone’s “Connect & Charge” tent. The festival’s massive organic social reach means that every selfie or live stream shared over the partner’s network effectively serves as implicit advertising.
Similarly, Glastonbury’s unique platform has also turned it into an important product and technology sandbox. In recent years, operators like Vodafone have turned heads with innovations like geofenced, time-limited eSIM trials and dedicated, custom 5G standalone (SA) network slices for payment terminals. The dense festival crowds create ideal conditions for testing nascent technologies whose value can be hard to demonstrate elsewhere.
Connecting the crowds at rural Worthy Farm remains an immense technical challenge
Glastonbury drops the population of a medium-sized city into a rural valley for five days. Unlike flagship events held in purpose-built and well-connected venues like Wimbledon or Silverstone, Worthy Farm is set in open countryside with little permanent fiber or power presence. As a result, everything from temporary cells on wheels (COWs) to ducts and generators needs to be transported down narrow lanes, approved by landowners, installed, and tested in the weeks leading up to the festival.
With more than 210,000 people circulating among more than 100 stages, secret sets and pop-up bars, the event exerts enormous pressure on mobile networks. Traffic hotspots shift constantly as the crowd moves, forcing engineers to repeatedly re-tune power levels and antenna tilt across sites. Vodafone anticipated record demand, projecting 270 TB of traffic on its network during this year’s festival, which is more than it typically carries in several midsized UK towns combined over a day.
Analysis of Speedtest Intelligence data reveals clusters of poor and very poor signal quality at Glastonbury 2025 tracked the locations with dense crowds, indicating congestion problems
To meet the challenge, each mobile operator once again deployed a network of COWs across Worthy Farm this year, adding dense capacity wherever crowds gathered. High-performing multi-band radios on the COWs provided both wide coverage (using bands like 700 and 800 MHz) and high capacity (using 3.5 GHz), with operators aggregating every last scrap of available spectrum to maximize network performance. Backhaul was mainly delivered through a mix of fiber and high-capacity microwave links operating in the E-Band (80 GHz).
The combination of high site density, with up to eleven COWs per operator, and the wide range of spectrum deployed, often spanning five or six carriers, means engineers have to carefully optimize for factors such as interference (when signals from different sites clash), inter-site distance (how close or far the equipment is placed), and TDD frame offsets (timing adjustments to keep data moving smoothly across the networks).
Three’s spectrum depth and EE’s spectrum diversity emerge as key performance differentiators at this year’s festival
Analysis of Speedtest Intelligence® data reveals that Three’s network delivered the highest median download speed at Worthy Farm this year (347.66 Mbps). It also exhibited the smallest gap between its pre-event baseline, when most tests were likely conducted by network engineers in ideal (limited load) conditions, and its in-event performance, highlighting the superior ability of its network to absorb the festival’s surge in demand.
Three leads in Median Download Speed at Glastonbury 2025, Reflecting its Mid-Band Advantage
Speedtest Intelligence® | 25–29 June 2025 vs. Preceding 30-Day Baseline
The operator’s lead in median download speed also translated into stronger results for quality of experience across core applications such as web browsing, video calling, and gaming. Users on its network, for example, saw faster page load times and lower latency to popular gaming servers on average.
Three Leads in Web Browsing, Video Calling and Gaming Performance During Glastonbury 2025
Speedtest Intelligence® | 25–29 June 2025
While Three may not have deployed the most temporary sites in raw numbers at the festival, as reflected by Vodafone leading in average 4G and 5G signal strength readings (a proxy for site density and low-band use), it was able to more than compensate for this by allocating a larger amount of spectrum per site and on a per-subscriber basis. This leverages Three’s inherent advantage of a contiguous 100 MHz block in the 3.5 GHz band, alongside its other valuable spectrum assets, and is helped by its smaller subscriber base (albeit one with a more demanding usage profile).
Although EE cannot match Three’s sheer mid-band spectrum depth and therefore trails in absolute peak (and median, here) download speeds at events like Glastonbury, its emphasis on spectrum diversity shines when conditions are less than ideal, such as at the cell edge or on congested sites. By deploying a wide range of bands across a broad frequency range, EE’s network achieved the best results across all key quality of service metrics, including download and upload speed and latency, at the 10th percentile during this year’s event. This demonstrates the advantage of distributing traffic over multiple carriers (improving device compatibility) and having deep low-band deployment.
EE Delivers Best Performance at the 10th Percentile at Glastonbury 2025
Speedtest Intelligence® | 25–29 June 2025 vs. 26–30 June 2024
While holding the vaunted “Connectivity Partner” title may have placed extra demands on Vodafone’s network at the festival, its infrastructure underpinned use cases like eSIM onboarding, free Wi-Fi at the “Connect & Charge” stand, and a dedicated 5G SA slice for payment terminals again this year. Despite these pressures, Vodafone delivered robust performance, achieving median download speeds of 106.74 Mbps during the event.
Vodafone’s performance at the 10th percentile, however, lagged behind all other networks at this year’s festival. As a result, its subscribers were more likely to encounter issues associated with lower speeds, such as video buffering, with Vodafone’s network recording 10th percentile download speeds less than half those of other operators. Analysis of signal strength and signal quality data indicates that its 5G layer was likely relatively more congested, while the opposite appeared true for its 4G network. This may reflect Vodafone’s spectrum mix, which forces heavier reliance on narrow low-band carriers for 5G, while its 4G network benefits from additional capacity through recently refarmed low and high-band 3G carriers.
Vodafone Leads on Signal Strength at Glastonbury 2025, Reflecting Superior Site Density and Low-Band Reach
Speedtest Intelligence® | 25–29 June 2025
Meanwhile, O2 trailed other operators in most performance metrics, providing lower median download speeds (67.66 Mbps) and poorer web page load times and latencies to popular gaming servers during the event. Its network also recorded poorer signal strength across 4G and 5G on average, potentially reflecting lower site density on Worthy Farm. Despite these comparative shortcomings, factors such as additional spectrum (made available through 3G refarming and the use of 700 MHz for 5G) enabled both Vodafone and O2 to significantly improve performance over the previous year’s festival, with median download speeds increasing by at least 25% for each.
O2's Poor Signal Quality on 4G Layer Indicates Congestion Problems
Speedtest Intelligence® | 25–29 June 2025
Glastonbury will return to a very different UK network landscape in 2027
Mobile operators will return to Worthy Farm in 2027 after a fallow year, equipped with significant network enhancements. They will have more spectrum to play with due to the Vodafone-Three merger and the upcoming mmWave auction, along with more advanced 5G SA configurations (with carrier aggregation extending to the uplink for the first time). New network demands, should they emerge (such as AI’s potential to increase the focus on latency), will shape deployments. In the interim, the battle for network leadership will continue to play out at the hundreds of other events happening across the UK this summer.
Ookla retains ownership of this article including all of the intellectual property rights, data, content graphs and analysis. This article may not be quoted, reproduced, distributed or published for any commercial purpose without prior consent. Members of the press and others using the findings in this article for non-commercial purposes are welcome to publicly share and link to report information with attribution to Ookla.
Luke Kehoe leads Ookla’s research and thought leadership efforts in Europe.
An electronic engineering alumnus of University College Dublin, Luke has extensive experience collaborating with mobile operators, telecoms vendors, and government agencies in research and advisory roles across Europe. He has contributed to internationally recognised thought leadership publications in areas such as 5G, IoT, open RAN, and edge computing, working with prestigious organisations like the Telecom Infra Project and the World Economic Forum.
Poland Races to Regain 5G Competitiveness in Europe with Mid-Band Rollout | Polska galopuje do odzyskania konkurencyjności 5G w Europie dzięki wdrożeniu średniego pasma częstotliwości
Poland’s operators are rapidly deploying mid-band 5G in an attempt to capture the growing premium market segment
Late to the game in staging a mid-band auction, Poland has lagged behind its European peers in 5G deployment in recent years. This delay has weighed on the country’s global competitiveness in mobile network performance and slowed its progress toward meeting the European Commission’s flagship 5G deployment targets, which require universal 5G coverage across every EU member state by the end of the decade. This article examines the state of Poland’s mobile market and its broader regional 5G competitiveness in the context of ongoing mid-band deployments. A follow-up report will assess the longer-term impact of the commercialization of the recently awarded low-band spectrum and ongoing network sunsets on network coverage and availability.
Key Takeaways:
Intensive capital spending on mid-band deployment drives substantial uplift in 5G performance across Polish operators from Q1 2024, pushing the country ahead of regional peers over the last year.Median 5G download speeds in Poland jumped by over 50% to 160.30 Mbps between Q1 2024 and Q1 2025, based on Speedtest Intelligence® data, propelling the country ahead of Czechia, Romania, and Slovakia for the first time in 5G performance. Despite this progress, Poland continues to trail its regional peers in 5G network Consistency, a measure of how reliably a mobile connection remains “fast enough” for normal use.
T-Mobile and Orange surpass Play and Plus in speed and select Quality of Experience (QoE) measures. Differences in how quickly and extensively Polish operators have deployed their mid-band spectrum assets have led to a diverging market profile since Q1 2024, with T-Mobile and Orange significantly extending their speed lead over their rivals. Between Q1 2024 and Q1 2025, median 5G download speeds rose by as much as 72% on Play (to 122.64 Mbps), 86% on T-Mobile (to 201.76 Mbps), and 90% on Orange (to 222.10 Mbps)—while declining by over 10% on Plus (to 116.76 Mbps).
Network investments have broadened 5G coverage in Poland, but significant regional disparities remain. Nationally, 5G availability rose from 28.5% in Q1 2024 to 43.1% in Q1 2025, driven by continued Dynamic Spectrum Sharing (DSS) rollouts and the activation of mid-band spectrum—placing the country ahead of regional peers Bulgaria, Romania, and Hungary in 5G availability. Nonetheless, by Q4 2024, a pronounced coverage gap persisted between the country’s best- and worst-served provinces, with 5G availability in the populous Masovian Voivodeship (47.2%) double that of the Lubusz Voivodeship (23.6%).
Over the last year, Polish operators have been locked in an intense four-way race to catch up with their regional peers in 5G deployment, driven by stringent coverage obligations imposed by the Polish telecoms regulator (UKE), a wave of funding support from Brussels, and a growing push to compete for a larger share of the country’s widening premium market segment, where network performance has emerged as a key competitive differentiator.
Poland’s mobile market is today awash with deployment activity, as operators ramp up capital spending to the highest levels in years to equip thousands of mobile sites with mid-band spectrum, accelerate the sunset of 3G networks, and lay the groundwork for launching 5G standalone (SA) in the coming years. This flurry of activity follows the completion of the 700/800 MHz auction at the end of March this year, where all Polish operators secured low-band 5G spectrum for the first time—paving the way for improved rural and deep in-building 5G coverage and rounding out the country’s 5G spectrum release plans.
While 5G capital spending has slowed across much of Europe, Poland sees different dynamics due to late spectrum auctions
Poland was notably late in releasing dedicated 5G spectrum in the ‘pioneer bands’ identified by the European Commission as critical to the timely commercialization and rollout of 5G across EU member states. The country’s mid-band (3.6 GHz) auction, initially planned for mid-2020, was repeatedly delayed—by more than three years—due to the pandemic and a protracted security legislation process.
These delays in spectrum availability have contributed to Poland’s divergence from much of the rest of Europe in both the economic and technical dimensions of the 5G rollout. Until recently, Polish mobile operators exhibited lower capital intensity (they invested less of their revenue) compared to peers in other European countries. Most of their spending went into upgrading 4G sites and preparing for the 3G shutdown, instead of building a new 5G mid-band capacity layer or expanding 5G coverage using low-band (700 MHz) spectrum.
Orange's Rising Mobile Capex Reflects 5G Network Expansion
Analysis of Orange Poland accounts | 2020 – 2024
Analysis of financial data published by Orange, Poland’s largest mobile operator by subscriber count, confirms that the era of lower capital intensity (relative to elsewhere in Europe) is over. The recent spectrum auctions have triggered a new cycle of investment, with Orange doubling its mobile network spending in the past three years. Play has also rapidly increased its investment, as its French parent Iliad reported injecting record amounts into Play’s mobile infrastructure last year.
Play's Contribution to Capex in the Iliad Group Surges as 5G Buildout Ramps Up
Analysis of Iliad Group accounts | 2020 – 2024
On the technical side, meanwhile, Poland’s spectrum delay meant that three of the country’s four operators were forced to rely heavily on Dynamic Spectrum Sharing (DSS)—a technology that allows 4G and 5G to operate on the same band and adjust ‘dynamically’ to demand—in an effort to deliver early 5G coverage in the 2100 MHz band while awaiting spectrum auctions. This strategy resulted in Poland’s initial 5G performance more closely resembling those typical of 4G networks, as DSS deployments are typically based on a 10 MHz carrier where part of the capacity is still reserved for 4G signals, making 5G speeds with DSS around 15–25 % lower than if the band were dedicated solely to 5G.
The limitations of using DSS to deliver a “5G experience” were exemplified by the speed advantage maintained by Plus earlier in the 5G rollout. Importantly, Plus was the only Polish operator that did not rely on DSS and instead dedicated a full 40 MHz carrier in the 2600 MHz (TDD) band to 5G before mid-band spectrum became available at the start of last year. Prior to the 3.5 GHz band coming online, when the other operators were still wholly dependent on DSS for 5G coverage, Plus’s median 5G download speed of 133.34 Mbps was as much as 77 % higher than T-Mobile’s, 81 % higher than Orange’s, and 92 % higher than Play’s.
Polish operators move from mid-band spectrum acquisition to mass commercial deployment in record time
The pent-up demand for mid-band spectrum in Poland was evident when mobile operators like Orange, T-Mobile, and Play launched commercial services just three months after acquiring mid-band spectrum, moving quickly from the auction in October 2023 to commercial launches by January 2024. T-Mobile reported that its mid-band 5G network already covered more than 25% of the Polish population by April 2024, with more than 2,100 sites active, while Orange announced it had reached 40% coverage by mid-June.
This rollout pace is exceptional by European standards and indicative of the increased pace of deployment possible later in the 5G technology cycle. It took Spain’s Telefónica (Movistar) about six months to reach its first 1,000 mid-band sites by comparison, and Germany’s operators needed around nine months to achieve the same milestone.
Plus's Spectrum Holdings in the 2600 MHz TDD Band Lend it a Decisive Capacity Lead
Each operator secured a contiguous 100 MHz block of spectrum in the 3.5 GHz band, which is widely regarded as optimal due to the large channel bandwidth this configuration affords. However, Plus has been notably slower to commercialise this allocation at scale. Plus’s earlier strategy of deploying 5G in the dedicated 2600 MHz band (rather than relying on DSS), alongside later using the 2100 MHz band as well, gave it more flexibility to delay a broad mid-band rollout as it previously enjoyed a significant 5G speed advantage over competitors while they were still heavily dependent on DSS deployments.
Mid-band deployment shifts 5G performance rankings among Polish operators
Mass deployment of a new capacity layer by the other three operators has since decisively altered performance dynamics in the Polish market and eroded Plus’s lead. In the space of one year between Q1 2024 and Q1 2025, Plus has moved from market leader in median 5G download speed to laggard, becoming the only Polish operator to see a year-on-year decline in 5G speed, down 10%, indicating the increasing limitations of its 2600 MHz strategy.
Orange and T-Mobile Pull Ahead in 5G Performance with Mid-Band Deployment
Speedtest Intelligence® | Q1 2023 – Q1 2025
By contrast, mid-band deployment has boosted performance across the rest of the market, with median 5G speeds rising by as much as 72% on Play, 86% on T-Mobile, and 90% on Orange between Q1 2024 and Q1 2025. While Orange led the Polish market in Q1 with a median 5G download speed of 222.11 Mbps, the operator’s lead has narrowed significantly as T-Mobile’s mid-band buildout has progressed, with T-Mobile now recording median 5G download speeds of 201.76 Mbps, well ahead of third- and fourth-placed Play (122.64 Mbps) and Plus (116.76 Mbps), respectively.
Plus's Lead in 5G Consistency Narrows as 2600 MHz Advantage Recedes with Mid-Band Deployment
Speedtest Intelligence® | Q1 2023 – Q1 2025
Despite losing its lead in median 5G download speed, Plus continues to lead at the 10th percentile (29.44 Mbps in Q1 2025), meaning subscribers in its lowest-performing areas still enjoy comparatively better speeds than those on rival networks. This advantage is likely linked to Plus’s lower dependence on DSS. However, T-Mobile (24.48 Mbps) and Orange (21.88 Mbps) are quickly closing the gap, with their 10th percentile 5G speeds now converging toward Plus. Plus’s 5G network consistency, measured as the proportion of Speedtest samples meeting a minimum download and upload threshold of 25/3 Mbps, has also declined over the past year, although it remains the market leader.
On upload performance, meanwhile, Play’s 5G network led the market in Q1 2025, recording median speeds of 19.33 Mbps, followed by Orange (18.99 Mbps), T-Mobile (17.32 Mbps), and Plus (14.96 Mbps). Unlike the substantial gains seen in download speeds, there is limited evidence so far that the mid-band rollout has materially improved upload performance, with median upload speeds about 6% lower in Q1 2025 compared to the same quarter last year. This discrepancy arises primarily because all four operators continue to deploy 5G in non-standalone (NSA) mode, requiring devices to transmit uplink traffic via existing 4G anchor bands. Consequently, the newly available 3.5 GHz spectrum enhances downlink capacity but leaves the congested 4G uplink path unchanged.
Play Develops Lead in 5G Upload Performance
Speedtest Intelligence® | Q1 2023 – Q1 2025
The operators’ investments in deploying a new 5G capacity layer have coincided with a broader RAN refresh effort, translating into improved quality of experience for users in key use cases such as video streaming and web browsing. Median web page load times on T-Mobile’s network, for instance, improved by around 4% between Q3 2024 and Q1 2025. Orange led in video metrics such as start time, resolution, and uninterrupted playback in the last quarter.
5G Drives QoE Improvements in Use Cases like Web Browsing
Speedtest Intelligence® | Q1 2025
Capital investment expands 5G coverage, but Poland’s rural-urban digital divide persists
While investments in DSS and the mid-band rollout have enabled Polish operators to make significant strides in 5G availability, which increased nationally from 28.5% in Q1 2024 to 43.1% in Q1 2025, regional coverage disparities continue to be a feature of the mobile network experience in Poland. Operators have prioritized 5G deployments in the richest and densest parts of Poland where fiber is heavily deployed, including the Masovian (Warsaw) and Pomeranian (Tri-City) provinces. In these provinces, 5G availability reached more than 40% by the end of last year and contributed to driving materially higher median download speeds than the national average.
5G Availability Remains Highly Varied Across Poland Outside of Urbanized Areas
Speedtest Intelligence® | 5G Availability (%) in Q4 2024
By contrast, border provinces along the south and west of the country continue to experience much lower levels of 5G availability. Lubusz had the lowest availability (23.6% at the end of last year), where there is lower population density and lower subscriber spending, which reduces operators’ commercial incentives for widespread 5G investment. This trend has driven the development of a notable speed gap between provinces, with mobile subscribers in Lubusz also experiencing the lowest median download speeds (59.97 Mbps) in Poland, almost 33% below the leading Masovian province.
Mobile Download Speeds Are Lower in Less Urbanized Areas of Poland
Speedtest Intelligence® | Median Download Speed (Mbps) in Q4 2024
Mid-band deployment improves Poland’s mobile competitiveness, but 5G consistency continues to trail regional peers
From a regional competitiveness lens, intensive mid-band deployments have been successful in breaking Poland’s cycle of mobile network underperformance, with median 5G download speeds rising by over 50% on average to 160.30 Mbps between Q1 2024 and Q1 2025. This has propelled the country ahead of Czechia, Romania, and Slovakia for the first time in terms of 5G download speed performance.
Despite Poland’s progress on its mid-band 5G rollout, the lingering effects of reliance on DSS and limited 5G spectrum diversity—up until the recent 700/800 MHz auction—mean that Poland continues to trail its regional peers in terms of 5G network consistency. In Q1 2025, 82% of Speedtest samples in Poland met the minimum 5G performance threshold for a consistent mobile experience, compared to 86% in Hungary, 89% in Romania, and 93% in Bulgaria.
Poland’s previous reliance on DSS, driven by limited 5G spectrum diversity, likely contributed to its slower average revenue per user (ARPU) growth compared to neighboring countries in recent years. Polish operators initially introduced tariffs with “5G at no extra cost” bolted onto existing 4G bundles, keeping prices flat to defend market share (and thereby maintaining depressed ARPU levels relative to regional peers). Combined with the external shock induced by markedly higher energy prices, stagnant ARPU levels created challenging operating conditions in the Polish market and weighed on operator profitability.
Intense Priced-Based Competition Precipitated Revenue Erosion in Poland During the First Half of the 5G Cycle
Analysis of GSMA Intelligence Data | % Change in Mobile ARPU (Q1 2020 vs Q1 2023)
In neighboring markets, by contrast, operators were able to leverage mid-band spectrum deployments as both technical and marketing levers, shifting their strategies from price competition toward service-based differentiation. This enabled them to more effectively upsell premium speed tiers or monetize specific use cases, such as fixed wireless access (FWA), which dedicated mid-band 5G deployments uniquely support.
T-Mobile and Play Outpaced Rivals in Subscription Share Growth in Recent Years
Analysis of UKE Market Data | 2019 – 2023
Similarly, the delayed timing of Poland’s mid-band 5G auction likely dampened supply-side factors key for driving growth in mobile data traffic. Between Q1 2020 and Q4 2024, traffic volumes in neighboring Bulgaria converged with that in Poland for the first time, increasing by 4.8x vs. Poland’s 2.6x. Meanwhile, Bulgarian operators capitalized early on mid-band spectrum availability to aggressively promote competitive FWA solutions (a major driver of mobile traffic in developed markets) and to introduce cheap unlimited data tariffs with fewer usage restrictions.
Poland Maintains Regional Lead in Mobile Data Volumes, but Bulgaria is Catching Up
Analysis of GSMA Intelligence data | 2020 – 2024
Polish operators have since sought to replicate Bulgaria’s success by debuting distinct marketing for their mid-band 5G deployments to differentiate the newer mid-band 5G rollouts from earlier DSS-based 5G networks in terms of performance and user experience. T-Mobile has leaned on ‘5G More’ branding, while Plus has used ‘5G Ultra’ to indicate the additional performance gains unlocked by their new 5G networks in locations where dedicated mid-band spectrum is deployed. This strategy has formed part of a broader shift in the market, with all operators moving away from a hyper-focus on price competition and toward ‘more for more’ pricing strategies, supporting improved profitability and renewed ARPU growth in the market with inflation-linked tariffs.
Poland Has Led Regional ARPU Growth Since Mid-Band 5G Deployments Started
Analysis of GSMA Intelligence Data | % Change in Mobile ARPU (Q1 2023 vs Q1 2025)
Low-band activation and network sunset progress set to reinforce mid-band 5G gains
With Poland’s telecom regulator, UKE, having set among Europe’s most ambitious coverage obligations for recent mid- and low-band spectrum auctions, operators are unlikely to delay commercial deployments in the newly acquired 700 and 800 MHz bands. These deployments are expected to start next month and will be crucial for establishing a national 5G coverage layer that, for the first time, extends deep indoors and into rural areas. This expanded coverage will also support wider rollout of voice over LTE (VoLTE) services, accelerating the 3G sunset and freeing up additional spectrum in the 900 MHz band.
We will revisit shortly to assess how Polish operators are progressing with deploying their new low-band spectrum and how effectively it is complementing the ongoing 3G sunset.
Polska galopuje do odzyskania konkurencyjności 5G w Europie dzięki wdrożeniu średniego pasma częstotliwości
Polscy operatorzy przyśpieszyli z wdrażaniem 5G w średnim paśmie, próbując przejąć rosnący segment rynku premium.
Polska, która spóźniła się z przeprowadzeniem aukcji na średnie pasmo, w ostatnich latach pozostawała w tyle za swoimi europejskimi rówieśnikami w zakresie wdrażania 5G. Opóźnienie to odbiło się na globalnej konkurencyjności kraju pod względem wydajności sieci mobilnych i spowolniło postępy w realizacji sztandarowych celów Komisji Europejskiej w zakresie wdrażania 5G, które wymagają powszechnego zasięgu 5G w każdym państwie członkowskim UE do końca dekady.
Niniejszy artykuł analizuje stan polskiego rynku telefonii komórkowej i jego szerszą regionalną konkurencyjność 5G w kontekście trwających wdrożeń średniego pasma. Kolejny raport oceni długoterminowy wpływ komercjalizacji niedawno przyznanego niskiego pasmana potrzeby pokryciowe 5G.
Kluczowe wnioski:
Intensywne wydatki kapitałowe na wdrożenie średniego pasma napędzają znaczny wzrost wydajności 5G u polskich operatorów od pierwszego kwartału 2024 r., pozycjonując kraj przed regionalnych konkurentów w ciągu ostatniego roku. Mediana prędkości pobierania 5G w Polsce wzrosła o ponad 50% do 160,30 Mb/s w okresie od I kwartału 2024 r. do I kwartału 2025 r., w oparciu o dane Speedtest Intelligence®, dzięki czemu Polska po raz pierwszy wyprzedziła Czechy, Rumunię i Słowację pod względem wydajności 5G. Pomimo tego postępu, Polska nadal pozostaje w tyle za swoimi regionalnymi rówieśnikami pod względem spójności sieci 5G, która jest miarą tego, jak niezawodnie zestawione połączenie mobilne pozostaje “wystarczająco szybkie” do normalnego użytkowania.
T-Mobile i Orange przewyższają Play i Plus pod względem prędkości i wybranych wskaźników jakości doświadczenia usług (QoE). Różnice w strategiach, jak szybko i szeroko polscy operatorzy wdrożyli swoje aktywa widma w średnim paśmie, doprowadziły do rozbieżnego profilu rynku od pierwszego kwartału 2024 r., przy czym T-Mobile i Orange znacznie zwiększyły swoją przewagę w zakresie prędkości nad rywalami. Pomiędzy I kwartałem 2024 r. a I kwartałem 2025 r. mediana prędkości pobierania 5G wzrosła aż o 72% w Play (do 122,64 Mb/s), 86% w T-Mobile (do 201,76 Mb/s) i 90% w Orange (do 222,10 Mb/s) – przy jednoczesnym spadku o ponad 10% w Plusie (do 116,76 Mb/s).
Inwestycje sieciowe zwiększyły zasięg 5G w Polsce, ale nadal utrzymują się znaczne różnice regionalne. W ujęciu krajowym dostępność sieci 5G wzrosła z 28,5% w I kwartale 2024 r. do 43,1% w I kwartale 2025 r., co wynikało z dalszego wdrażania dynamicznego współdzielenia widma (DSS) i aktywacji widma w średnim paśmie, dzięki czemu Polska wyprzedziła pod względem dostępności sieci 5G regionalne kraje takie jak Bułgaria, Rumunia i Węgry. Niemniej jednak do IV kwartału 2024 r. utrzymywała się wyraźna luka w zasięgu między najlepiej i najgorzej obsługiwanymi województwami w kraju, przy czym dostępność 5G w zaludnionym województwie mazowieckim (47,2%) była dwukrotnie wyższa niż w województwie lubuskim (23,6%).
Wyłączenia sieci 3G (ang. “3G sunset”) powodują gwałtowny spadek czasu spędzonego na 3G w 2024 r., ponieważ polscy operatorzy reorganizują widmo dla 4G (ang. “refarming”), ale ma to ogromny wpływ na dostępność usług w miejscach mniej zurbanizowanych.Podczas gdy T-Mobile pozostał jedynym polskim operatorem, który w pełni zakończył proces wygaszania sieci 3G do pierwszego kwartału 2025 r., zarówno Orange, jak i Play czynią obecnie znaczne postępy w zakresie refarmingu widma 3G 900 MHz i 2100 MHz na potrzeby 4G. Czas spędzony na 3G spadł poniżej 3% dla obu operatorów do końca 2024 roku. Natomiast abonenci Plusa nadal spędzali znacznie więcej czasu w sieci 3G – 13,41% na koniec 2024 roku.
W ciągu ostatniego roku polscy operatorzy byli jednak zamknięci w intensywnymwyścigu, aby dogonić swoich regionalnych kolegów we wdrażaniu 5G, napędzanym przez rygorystyczne obowiązki w zakresie zasięgu nałożone przez polskiego regulatora telekomunikacyjnego (UKE), falę wsparcia finansowego z Brukseli i rosnące dążenie do konkurowania o większy udział w poszerzającym się segmencie rynku premium w kraju, w którym wydajność sieci stała się kluczowym wyróżnikiem konkurencyjnym.
Polski rynek telefonii komórkowej jest dziś zdominowany aktywnością wdrożeniową, stąd operatorzy zwiększają wydatki kapitałowe do najwyższych poziomów od lat, aby wyposażyć tysiące stacji bazowych w widmo średniego pasma, przyspieszyć wyłączanie sieci 3G i położyć podwaliny pod uruchomienie samodzielnej sieci 5G (SA) w nadchodzących latach. Taką falę aktywności można zwłaszcza zauważyć po zakończeniu aukcji 700/800 MHz pod koniec marca tego roku, w której wszyscy polscy operatorzy po raz pierwszy zabezpieczyli widmo 5G w niskim paśmie – torując sobie drogę do poprawy zasięgu 5G na obszarach wiejskich i głęboko wewnątrz budynków (ang. “deep in-building”) w miastach oraz uzupełniając krajowe plany udostępniania widma 5G.
Podczas gdy wydatki kapitałowe na 5G spowolniły w dużej części Europy, Polska doświadcza inną dynamikę ze względu na późne aukcje na częstotliwości
Polska znacznie spóźniła się z udostępnieniem dedykowanych częstotliwości 5G w “pionierskich” pasmach zidentyfikowanych przez Komisję Europejską jako krytyczne dla terminowej komercjalizacji i wdrożenia 5G w państwach członkowskich UE. Krajowa aukcja częstotliwości pasma środkowego (3,6 GHz), początkowo planowana na połowę 2020 r., była wielokrotnie opóźniona – o ponad trzy lata – z powodu pandemii i przedłużającego się procesu legislacyjnego w zakresie bezpieczeństwa.
Te opóźnienia w dostępności częstotliwości przyczyniły się do tego, że Polska odbiega od reszty Europy zarówno w wymiarze ekonomicznym, jak i technicznym wdrażania 5G. Do niedawna polscy operatorzy komórkowi wykazywali niższą kapitałochłonność (inwestowali mniejszą część swoich przychodów) w porównaniu do innych europejskich operatorów. Większość ich wydatków przeznaczono na modernizację 4G i przygotowanie do wyłączenia 3G, zamiast budować nową warstwę pojemności 5G w średnim paśmie lub rozszerzać zasięg 5G przy użyciu niskich częstotliwości (700 MHz).
Rosnące nakłady Orange na sieć mobilną odzwierciedlają rozwój sieci 5G
Analiza rachunków Orange Polska | 2020–2024
Analiza danych finansowych opublikowanych przez Orange, największego operatora komórkowego w Polsce pod względem liczby abonentów, potwierdza, że era niższej kapitałochłonności (w porównaniu z innymi krajami w Europie) dobiegła końca. Niedawne aukcje częstotliwości wywołały nowy cykl inwestycyjny, a Orange podwoił wydatki na sieć mobilną w ciągu ostatnich trzech lat. Play również gwałtownie zwiększył swoje inwestycje, jego francuska spółka dominująca Iliad poinformowała w zeszłym roku o zainwestowaniu rekordowych kwot w infrastrukturę mobilną Play.
Udział Play w nakładach inwestycyjnych Grupy Iliad gwałtownie rośnie wraz z przyspieszeniem rozbudowy sieci 5G
Analiza rachunków Grupy Iliad | 2020–2024
Tymczasem od strony technicznej opóźnienie aukcji częstotliwości 5G w Polsce oznaczało, że trzech z czterech operatorów w kraju było zmuszonych w dużym stopniu polegać na dynamicznym współdzieleniu widma (ang. “Dynamic Spectrum Sharing” – DSS) – technologii, która pozwala 4G i 5G działać w tym samym paśmie i “dynamicznie” dostosowywać się do zapotrzebowania na pojemność danej technologii – w celu zapewnienia wczesnego zasięgu 5G w paśmie 2100 MHz w oczekiwaniu na aukcje częstotliwości. Strategia ta spowodowała, że początkowa wydajność 5G w Polsce bardziej przypominała typową dla sieci 4G, ponieważ wdrożenia DSS są zwykle oparte na nośnej 10 MHz, w której część pojemności jest nadal zarezerwowana dla sygnałów 4G, co powoduje, że prędkości 5G z DSS są o około 15-25% niższe niż gdyby pasmo było przeznaczone wyłącznie dla 5G.
Ograniczenia wykorzystania DSS do zapewnienia “doświadczenia 5G” zostały zilustrowane przewagą prędkości utrzymywaną przez Plusa na wcześniejszym etapie wdrażania 5G. Co ważne, Plus był jedynym polskim operatorem, który nie polegał na DSS i zamiast tego przeznaczył pełną nośną 40 MHz w paśmie 2600 MHz (TDD) na 5G, zanim na początku ubiegłego roku częstotliwości średniego pasma stały się dostępne. Przed uruchomieniem pasma 3,5 GHz, gdy pozostali operatorzy byli nadal w pełni zależni od DSS w zakresie zasięgu 5G, średnia prędkość pobierania 5G Plusa wynosząca 133,34 Mb/s była aż o 77% wyższa niż w T-Mobile, 81% wyższa niż w Orange i 92% wyższa niż w Play.
Intensywne wdrażanie średniego pasma podnosi regionalną konkurencyjność Polski w zakresie 5G i zmienia dynamikę operatorów
Polscy operatorzy w rekordowym czasie przechodzą od zakupu częstotliwości w średnim paśmie do masowego wdrożenia komercyjnego
Stłumiony popyt na częstotliwości średniego pasma w Polsce był widoczny, gdy operatorzy komórkowi, tacy jak Orange, T-Mobile i Play, uruchomili usługi komercyjne zaledwie trzy miesiące po nabyciu częstotliwości średniego pasma, szybko przechodząc od aukcji w październiku 2023 r. do komercyjnego uruchomienia do stycznia 2024 roku. T-Mobile poinformował, że jego średniopasmowa sieć 5G obejmowała już ponad 25% populacji Polski do kwietnia 2024 r., z ponad 2100 aktywnymi stacjami bazowymi, podczas gdy Orange ogłosił, że osiągnął 40% zasięgu do połowy czerwca.
To tempo wdrażania jest wyjątkowe jak na standardy europejskie i wskazuje na zwiększone tempo wdrażania możliwe w późniejszym okresie cyklu technologicznego 5G. Dla porównania, hiszpańska Telefónica (Movistar) potrzebowała około sześciu miesięcy, aby osiągnąć pierwsze 1000 stacji bazowych w średnim paśmie, a niemieccy operatorzy potrzebowali około dziewięciu miesięcy, aby osiągnąć ten sam kamień milowy.
Zasoby częstotliwości Plus w paśmie 2600 MHz TDD zapewniają mu zdecydowaną przewagę przepustowości
Każdy z operatorów zabezpieczył ciągły blok częstotliwości o szerokości 100 MHz w paśmie 3,5 GHz, który jest powszechnie wykorzystywany. Jednak Plus był znacznie wolniejszy w komercjalizacji tej alokacji na dużą skalę. Wcześniejsza strategia Plusa polegająca na wdrażaniu 5G w dedykowanym paśmie 2600 MHz (zamiast polegać na DSS), a później także na wykorzystaniu pasma 2100 MHz, dała mu większą elastyczność w opóźnianiu szerokiego wdrożenia średniego pasma, ponieważ wcześniej cieszył się znaczną przewagą prędkości 5G nad konkurentami, podczas gdy byli oni nadal silnie uzależnieni od wdrożeń DSS.
Wdrożenie średniego pasma zmienia rankingi wydajności 5G wśród polskich operatorów
Masowe wdrożenie nowej warstwy pojemności przez pozostałych trzech operatorów zdecydowanie zmieniło dynamikę wydajności 5G na polskim rynku i zmniejszyło przewagę Plusa. W ciągu jednego roku, między pierwszym kwartałem 2024 r. a pierwszym kwartałem 2025 r., Plus przesunął się z lidera rynku pod względem mediany prędkości pobierania 5G do jednego z wolniejszych, stając się jedynym polskim operatorem, który odnotował spadek prędkości 5G rok do roku, o 10%, co wskazuje na rosnące ograniczenia jego strategii 2600 MHz.
Orange i T-Mobile zyskują przewagę w wydajności 5G dzięki wdrożeniu pasma średniego
Speedtest Intelligence® | I kwartał 2023 – I kwartał 2025
Z kolei wdrożenie średniego pasma zwiększyło wydajność na pozostałej części rynku, a mediana prędkości 5G wzrosła aż o 72% w Play, 86% w T-Mobile i 90% w Orange między 1. kwartałem 2024 r. a 1. kwartałem 2025 r. Podczas gdy Orange był liderem polskiego rynku w pierwszym kwartale ze średnią prędkością pobierania 5G wynoszącą 222,11 Mb/s, przewaga operatora znacznie się zmniejszyła wraz z postępem budowy średniego pasma T-Mobile, przy czym T-Mobile odnotowuje obecnie medianę prędkości pobierania 5G na poziomie 201,76 Mb/s, znacznie wyprzedzając odpowiednio trzeciego i czwartego Play (122,64 Mb/s) i Plusa (116,76 Mb/s).
Przewaga Plusa w spójności 5G maleje, gdy przewaga pasma 2600 MHz ustępuje wraz z wdrożeniem pasma średniego
Speedtest Intelligence® | I kwartał 2023 – I kwartał 2025
Pomimo utraty pozycji lidera pod względem mediany prędkości pobierania 5G, Plus nadal prowadzi w 10. percentylu (29,44 Mb/s w 1. kwartale 2025 r.), co oznacza, że abonenci w obszarach o najniższych wynikach nadal cieszą się stosunkowo lepszymi prędkościami niż abonenci konkurencyjnych sieci. Przewaga ta jest prawdopodobnie związana z mniejszą zależnością Plusa od DSS. Jednak T-Mobile (24,48 Mb/s) i Orange (21,88 Mb/s) szybko zmniejszają lukę, a ich 10-procentowe prędkości 5G zbliżają się teraz do Plusa. Spójność sieci 5G Plusa, mierzona jako odsetek próbek Speedtest spełniających minimalny próg pobierania i wysyłania 25/3 Mbps, również spadła w ciągu ostatniego roku, chociaż pozostaje liderem rynku.
Tymczasem pod względem wydajności wysyłania, sieć 5G Play była liderem na rynku w pierwszym kwartale 2025 r., odnotowując medianę prędkości 19,33 Mb/s, a następnie Orange (18,99 Mb/s), T-Mobile (17,32 Mb/s) i Plus (14,96 Mb/s).
W przeciwieństwie do znacznych wzrostów prędkości pobierania, jak dotąd istnieją ograniczone dowody na to, że wdrożenie średniego pasma znacznie poprawiło wydajność wysyłania, przy czym mediana prędkości wysyłania była o około 6% niższa w pierwszym kwartale 2025 r. w porównaniu z tym samym kwartałem ubiegłego roku. Rozbieżność ta wynika przede wszystkim z faktu, że wszyscy czterej operatorzy nadal wdrażają 5G w trybie non-standalone (NSA), nadal wymagają od urządzeń technologii 4G do obsługi ruchu wysyłania i warstwy sygnałowej. W związku z tym nowo dostępne widmo 3,5 GHz zwiększa przepustowość łącza w dół, ale pozostawia zatłoczoną ścieżkę łącza 4G w górę bez zmian.
Play zyskuje przewagę w wydajności wysyłania danych w sieci 5G
Speedtest Intelligence® | I kwartał 2023 – I kwartał 2025
Inwestycje operatorów we wdrażanie nowej warstwy przepustowości 5G zbiegły się w czasie z szerszymi działaniami w zakresie modernizacji sieci RAN, przekładając się na lepszą jakość usług doświadczanych przez użytkowników w kluczowych zastosowaniach, takich jak wideo streaming i przeglądanie stron internetowych. Na przykład mediana czasu ładowania strony internetowej w sieci T-Mobile poprawiła się o około 4% między 3. kwartałem 2024 r. a 1. kwartałem 2025 r., co stawia ją w czołówce pod tym względem. Tymczasem Orange był liderem pod względem wskaźników wideo, takich jak czas rozpoczęcia, rozdzielczość i nieprzerwane odtwarzanie w ostatnim kwartale.
5G napędza poprawę jakości doświadczeń (QoE) w zastosowaniach takich jak przeglądanie stron internetowych
Speedtest Intelligence® | I kwartał 2025
Inwestycje kapitałowe zwiększają zasięg 5G, ale przepaść cyfrowa między wsią a miastem w Polsce utrzymuje się
Podczas gdy inwestycje w DSS i wdrożenie średniego pasma umożliwiły polskim operatorom poczynienie znaczących postępów w zakresie dostępności 5G, która wzrosła w skali kraju z 28,5% w I kwartale 2024 r. do 43,1% w I kwartale 2025 r., regionalne różnice w zasięgu nadal są cechą charakterystyczną sieci mobilnej w Polsce.
Operatorzy nadali priorytet wdrożeniom 5G w najbogatszych i najbardziej zaludnionych częściach Polski, gdzie światłowody są mocno rozwinięte, w tym w województwach mazowieckim (Warszawa) i pomorskim (Trójmiasto). W tych województwach dostępność 5G osiągnęła ponad 40% pod koniec ubiegłego roku i przyczyniła się do osiągnięcia znacznie wyższych średnich prędkości pobierania niż średnia krajowa.
Dostępność 5G pozostaje wysoce zróżnicowana w Polsce poza obszarami zurbanizowanymi
Speedtest Intelligence® | Dostępność 5G (%) w IV kw. 2024
Natomiast województwa przygraniczne na południu i zachodzie kraju nadal doświadczają znacznie niższych poziomów dostępności 5G. Województwo lubuskie miało najniższą dostępność (23,6% na koniec ubiegłego roku), gdzie występuje mniejsza gęstość zaludnienia i niższe wydatki abonentów, co zmniejsza zachęty komercyjne operatorów do powszechnych inwestycji w 5G. Tendencja ta doprowadziła do powstania znacznej luki prędkości między województwami, a abonenci mobilni w Lubuskiem również doświadczają najniższej mediany prędkości pobierania (59,97 Mb/s) w Polsce, prawie 33% poniżej wiodącego województwa mazowieckiego.
Prędkości pobierania w sieciach mobilnych są niższe na mniej zurbanizowanych obszarach Polski
Speedtest Intelligence® | Mediana prędkości pobierania (Mbps) w IV kw. 2024
Wdrożenie średniego pasma poprawia konkurencyjność mobilną Polski, ale spójność 5G nadal ustępuje regionalnym konkurentom
Z punktu widzenia konkurencyjności regionalnej, intensywne wdrożenia średniego pasma skutecznie przełamały cykl słabej wydajności sieci mobilnej w Polsce, a mediana prędkości pobierania 5G wzrosła średnio o ponad 50% do 160,30 Mb/s między 1. kwartałem 2024 r. a 1. kwartałem 2025 r. Dzięki temu Polska po raz pierwszy wyprzedziła Czechy, Rumunię i Słowację pod względem prędkości pobierania 5G.
Pomimo postępów Polski we wdrażaniu 5G w średnim paśmie, utrzymujące się skutki polegania na DSS i ograniczonej różnorodności widma 5G aż do niedawnej aukcji 700/800 MHz oznaczają, że Polska nadal pozostaje w tyle za swoimi regionalnymi rówieśnikami pod względem spójności sieci 5G. W pierwszym kwartale 2025 r. 82% próbek Speedtest w Polsce spełniło minimalny próg wydajności 5G dla spójnego doświadczenia mobilnego, w porównaniu do 86% na Węgrzech, 89% w Rumunii i 93% w Bułgarii.
Nowo pozyskana różnorodność częstotliwości 5G daje polskim operatorom potężne narzędzie do stymulowania wzrostu ARPU
Wcześniejsza zależność Polski od DSS, wynikająca z ograniczonej różnorodności widma 5G, prawdopodobnie przyczyniła się do wolniejszego wzrostu średniego przychodu na użytkownika (ARPU) w porównaniu z sąsiednimi krajami na przestrzeni ostatnich lat. Polscy operatorzy początkowo wprowadzili taryfy z “5G bez dodatkowych kosztów” dodane do istniejących pakietów 4G, utrzymując ceny na stałym poziomie w celu obrony udziału w rynku (a tym samym utrzymując obniżone poziomy ARPU w porównaniu do regionalnych konkurentów). W połączeniu z zewnętrznym szokiem makroekonomicznym wywołanym znacznie wyższymi cenami energii, stagnacja poziomów ARPU stworzyła trudne warunki operacyjne na polskim rynku i wpłynęła na rentowność operatorów.
Intensywna konkurencja cenowa spowodowała erozję przychodów w Polsce w pierwszej połowie cyklu 5G
Analiza danych GSMA Intelligence | Zmiana procentowa ARPU w usługach mobilnych (I kw. 2020 vs I kw. 2023)
Z kolei na sąsiednich rynkach operatorzy byli w stanie wykorzystać wdrożenie częstotliwości w średnim paśmie zarówno jako korzyści techniczne, jak i marketingowe, przenosząc swoje strategie z konkurencji cenowej na zróżnicowanie oparte na usługach. Pozwoliło im to skuteczniej sprzedawać wyższe poziomy prędkości lub zarabiać na konkretnych rozwiązaniach, takich jak stały dostęp bezprzewodowy (FWA), dla którego działania wdrożone 5G w średnim paśmie nadaje się idealnie.
T-Mobile i Play wyprzedziły konkurentów w tempie wzrostu udziału subskrypcji w ostatnich latach
Analiza danych rynkowych UKE | 2019–2023
Podobnie, opóźniony termin polskiej aukcji 5G dla średniego pasma prawdopodobnie osłabił czynniki po stronie podaży, będące kluczowymi dla napędzania wzrostu konsumpcji danych z sieci mobilnych. W okresie od I kwartału 2020 r. do IV kwartału 2024 r. wolumen ruchu w sąsiedniej Bułgarii po raz pierwszy zrównał się z wolumenem w Polsce, wzrastając 4,8-krotnie w porównaniu do 2,6-krotnego wzrostu w Polsce.
W międzyczasie bułgarscy operatorzy wcześnie wykorzystali dostępność widma w średnim paśmie, aby agresywnie promować konkurencyjne rozwiązania FWA (główny czynnik napędzający ruch mobilny na rynkach rozwiniętych) i wprowadzić tanie taryfy nieograniczonej transmisji danych z mniejszymi ograniczeniami użytkowania.
Polska utrzymuje regionalne prowadzenie w wolumenach danych mobilnych, ale Bułgaria szybko nadrabia
Analiza danych GSMA Intelligence | 2020–2024
Od tego czasu polscy operatorzy starali się powtórzyć sukces Bułgarii, wprowadzając odrębny marketing dla swoich wdrożeń 5G w średnim paśmie, aby odróżnić nowsze wdrożenia 5G w średnim paśmie od wcześniejszych. T-Mobile oparł się na marce “5G Bardziej”, podczas gdy Plus użył sloganu marketingowego “5G Ultra”, aby wskazać dodatkowy wzrost wydajności odblokowany przez ich nowe sieci 5G w lokalizacjach, w których wdrożono dedykowane częstotliwości średniego pasma. Strategia ta stała się częścią szerszej zmiany na rynku, w której wszyscy operatorzy odchodzą od hiper-koncentracji opierającej się na konkurencji cenowej w kierunku strategii cenowych “więcej za więcej”, wspierając poprawę rentowności i ponowny wzrost ARPU.
Polska przoduje w regionalnym wzroście ARPU od momentu rozpoczęcia wdrożeń średniego pasma 5G
Analiza danych GSMA Intelligence | Zmiana procentowa ARPU w usługach mobilnych (I kw. 2023 vs I kw. 2025)
Aktywacja niskiego pasma i postępy w budowie sieci mają na celu wzmocnienie zysków 5G w średnim paśmie
W związku z tym, że polski regulator telekomunikacyjny, UKE, ustanowił jeden z najbardziej ambitnych zobowiązań dotyczących zasięgu w Europie dla ostatnich aukcji częstotliwości średniego i niskiego pasma, operatorzy raczej nie opóźnią komercyjnych wdrożeń w nowo nabytych pasmach 700 i 800 MHz. Oczekuje się, że wdrożenia te rozpoczną się w przyszłym miesiącu i będą miały kluczowe znaczenie dla ustanowienia krajowej warstwy zasięgu 5G, która znacznie poprawi pokrycie ciężko dostępnych miejsc wewnątrz budynków w miastach i zdalnych obszarów wiejskich. Rozszerzony zasięg będzie również wspierał szersze wdrażanie usług głosowych przez LTE (VoLTE), przyspieszając schyłek 3G i uwalniając dodatkowe widmo w paśmie 900 MHz.
Wkrótce powrócimy do tego tematu, aby ocenić, jak polscy operatorzy radzą sobie z wdrażaniem nowych częstotliwości niskopasmowych i jak skutecznie uzupełniają trwający proces wygaszania 3G.
Ookla retains ownership of this article including all of the intellectual property rights, data, content graphs and analysis. This article may not be quoted, reproduced, distributed or published for any commercial purpose without prior consent. Members of the press and others using the findings in this article for non-commercial purposes are welcome to publicly share and link to report information with attribution to Ookla.
Luke Kehoe leads Ookla’s research and thought leadership efforts in Europe.
An electronic engineering alumnus of University College Dublin, Luke has extensive experience collaborating with mobile operators, telecoms vendors, and government agencies in research and advisory roles across Europe. He has contributed to internationally recognised thought leadership publications in areas such as 5G, IoT, open RAN, and edge computing, working with prestigious organisations like the Telecom Infra Project and the World Economic Forum.
How Spain's Mobile Networks Performed During the Iberian Grid Collapse [Visualized] | Cómo Se Comportaron Las Redes Móviles Españolas Durante El Colapso De La Red Ibérica
Variation in outage scale and duration across mobile operators during the April 28th event reinforces the case that robust power resilience is now the single most decisive factor in preserving service continuity in mobile networks
The Iberian grid collapse stands as the starkest demonstration yet of the fragility of mobile network infrastructure precisely when it is needed most. Public scrutiny following the historic event has centred on understanding and ultimately mitigating the societal vulnerabilities stemming from mobile networks’ inherent reliance on a constant grid power supply, which is increasingly difficult to guarantee as severe events become more frequent.
For the first time, and with the goal of fostering deeper understanding of the interdependence between mobile and power infrastructure in Europe, Ookla is publishing detailed data exposing the scale and geographic distribution of mobile network disruptions caused by the April 28th event. Building on our earlier analysis, this research reveals that the impact of the power outage on mobile network infrastructure varied significantly across operators, with differing levels of power backup penetration likely a critical factor shaping the severity experienced by end users.
Key Takeaways:
At the peak of the April 28th blackout, more than half of all mobile network users across large parts of Spain were left completely without mobile signal. The proportion of mobile network users experiencing complete service loss (unable to call, text, or use data due to sites going dark) surged from a pre-blackout baseline of under 0.5% to more than 50% across extensive areas of Spain at the peak late on April 28th. This indicates a severe, widespread, and historic collapse of the mobile site grid that deepened throughout the afternoon and evening on the day of the blackout as available power backups were progressively depleted.
The timing and distribution of mobile network outages closely tracked the pattern of power grid events, underlining the telecom infrastructure’s vulnerability to even brief interruptions in power supply. Within 30 minutes of the grid failure (by 13:00 CET), the proportion of users with no service surged as small cells and sites with minimal power backup and battery autonomy went offline. After two hours, about 12% of users on the most affected operator had no service. Outage growth then slowed, suggesting that remaining macro sites, likely equipped with four-to-six hour battery banks, kept operating until their reserves ran out, leading to a final sharp wave of service loss in the late evening. The restoration of mobile services tracked the geographically-phased re-energisation of the power grid, with network disruptions persisted longer into the night in parts of Andalusia and Galicia.
While severe network outages were observed across all Spanish operators during the blackout, mobile users on the Vodafone network were less likely to experience a complete service loss. Four-to-eight hours after the grid collapse, the interval in which every operator hit its worst point in terms of service loss, Vodafone’s subscribers were, on average, less than half as likely to be left without service as subscribers on Orange’s network and notably less likely than subscribers on Movistar or Yoigo as well.
Morocco’s mobile site footprint remained operational throughout April 28th since domestic grid supply was unaffected. However, the country’s reliance on Spain for international connectivity in deeper network layers resulted in cascading failures and severe service degradation. The proportion of mobile users experiencing no service on the Orange Maroc and Maroc Telecom networks remained consistent with the pre-blackout baseline on April 28th, confirming there were no sustained network disruptions at the mobile site level due to grid failure, unlike in Spain and Portugal. However, analysis of Speedtest Intelligence data reveals significant performance degradation still occurred in Morocco, with the median load time for popular websites increasing by more than 20% compared to the same day in the previous week. The quality of experience (QoE) for Orange Maroc subscribers was particularly impacted during the blackout, reportedly due to disruptions in upstream subsea connectivity between Morocco and mainland Spain.
Power Resilience and Energy Management Strategies Shape the Anatomy of Network Outages
The Iberian grid collapse exemplifies a type of stress test becoming increasingly common in Europe—the ability of mobile network infrastructure to withstand prolonged, severe external shocks beyond direct operator, ensuring continuity of service precisely when it is most critical for public safety and societal functioning.
Beyond the April 28th blackout, recent events in the UK and Ireland, where winter storms (especially Storm Éowyn) caused extensive localized damage to electricity distribution networks, and in France, where substation vandalism resulted in brief but severe blackouts in Cannes and Nice, highlight electricity supply disruptions as a principal vulnerability for mobile networks reliant entirely on grid power. These disruptions add to other external vulnerabilities faced by operators, such as terrestrial or subsea fiber connectivity, upstream cloud links, and third-party peering connections.
Power resilience, particularly through battery and generator backup systems at mobile sites, has emerged as a key proactive measure to mitigate network outages resulting from electricity grid disruptions. Simply put, backup power serves a role analogous to public health measures in a pandemic: it strategically delays the onset of outages (extending the operational hours before sites lose power) and reduces the peak severity (limiting the total number of sites simultaneously affected).
Capital investments in network hardening tools like power redundancy can therefore “flatten” the service-impact curve during an outage event in the same way public health measures flattened the infection curve during the coronavirus pandemic. Prolonged and wide-area grid disruptions like the one on April 28th, however, demonstrate that no single measure is a silver bullet and long-term power autonomy is often not economically viable across a large proportion of the site footprint.
Recognizing this, mobile operators typically implement aggressive energy management measures during power disruptions to maximize site uptime and strategically allocate network resources based on user priority. These measures may include throttling site transmit power to reduce coverage footprints, limiting spectrum diversity to limit carrier aggregation and overall capacity, and temporarily disabling newer technologies such as 5G and Massive MIMO to extend battery and generator runtime.
Differences in the extent of power backup deployment and the strategic use of energy conservation and load-shedding measures shape the anatomy of network outages. Such variations among operators may arise from a diversity of fuel choices for backup power (e.g., batteries might offer long-term monetization potential, including opportunities for resale to the grid, but typically have shorter runtimes), network configurations (such as RAN sharing arrangements and site types, with dense urban small cells facing greater physical constraints for power backup deployments), and subscriber base characteristics (e.g., operators serving a larger rural subscriber base face more complex and costly challenges in enhancing network resilience).
Iberian Grid Collapse Cascaded through Mobile Networks in Spain, Moving in Lockstep with Power Disruptions
Analysis of Ookla® background signal scan data reveals that the April 28th event placed unprecedented stress on mobile networks in Spain, triggering a rapid collapse in site grid density. The historic scale of this collapse severely curtailed the coverage footprint of operator infrastructure, pushing a significant share of Spanish mobile subscribers into a ‘no service’ state, unable to connect to a nearby mobile site and thus temporarily unable to make calls, send texts, or use data.
Geospatial sequencing of the ‘no service’ data calculates the average probability that a mobile network user was left without signal during any given time window and serves as a proxy for the overall proportion of subscribers left with no network access. This methodology illustrates that the impact of the power blackout quickly cascaded through Spain’s mobile networks (closely mirroring the infection curve analogy described earlier):
Baseline (Pre-Blackout): Mobile sites were operating as normal. The average proportion of subscribers without service was less than 0.5%, reflecting the very high levels of network availability typical of Spain.
Initial Impact (Immediate Aftermath of Blackout): The grid collapsed at ~12:33 CET across the Iberian Peninsula, triggering rectifiers or uninterruptible power supply (UPS) devices at mobile sites to switch over to backup batteries or generators (where these were installed and adequately charged or fueled).
Within 30 minutes of the voltage drop, the proportion of subscribers without service had climbed well above the pre-event baseline across Spain and the outage curve entered a phase of runaway growth. This limited initial outage impact likely reflected the immediate loss of a small portion of the overall site footprint in areas where power backup was either absent or severely limited in capacity—most commonly at small cell sites in dense urban environments, where physical space restricts battery or generator installations, or at remote rural sites lacking redundancy (or featuring generators that took longer to switch over and come online).
Notably, the balanced geographical spread of subscribers beginning to experience no service on their devices (i.e. losing network access) at the start of the network outages suggests that the timing of impacts was broadly consistent across Spain, affecting both rural and urban regions. However, the initial severity was more pronounced in areas along the east and south coasts, as well as in the interior of the country, potentially indicating a lower level of power autonomy at sites in these areas.
Ramp-Up (2-8 Hours After Blackout): Within two hours of the blackout, the most affected operator had already seen ~12% of users left with no service. After this point, the rate of outage growth slowed temporarily for several hours. This pattern suggests that the power backup profile of Spanish mobile sites was bi- or tri-modal, with a tiered approach to backup capacity employed by operators (i.e., site autonomy was clustered into configurations like two, four or six-hour power resilience, with battery discharge rates determined by amp-hour capacity and site load at each location).
As a result of this tiered approach to power autonomy and site load variability, a portion of sites continued operating, likely in a reduced state with aggressive energy conservation measures in place (reflected in marked performance degradation observed in Speedtest Intelligence® data), throughout the afternoon and evening of April 28th.
The eventual depletion of larger power reserves, typically found at major macro sites on lattice or monopole towers, triggered a final sharp wave of network outages four to eight hours after the blackout began. The loss of these wide-area ‘umbrella’ sites in the evening resulted in a substantial share of Spanish mobile subscribers experiencing no service by 21:00 CET, with severe network outages geographically distributed across all of Spain.
Peak (8-10 Hours After Blackout): The impact of the blackout on mobile network infrastructure in Spain peaked between 21:00 and 22:00 CET, with more than half of all mobile subscribers left without service across many regions by late evening, just before power began to return. By this stage (about ten hours after the initial power loss), even the most capable battery backups were likely exhausted, leaving only those sites powered by mobile or stationary generators still operational (especially where strategic fuel deliveries could extend uptime).
Recovery (Remainder of April 28th and 29th): Cross-analysis of geospatial sequencing of the no service data with nighttime light emissions detailed in NASA satellite imagery reveals that the recovery of mobile services and the site grid footprint in Spain closely tracked the pattern of grid re-energization. The proportion of users without service began to decline in the mid-afternoon and early evening in the areas where power was first restored, as the grid operator carried out a phased nationwide black-start. This early network recovery was most evident in regions such as the Basque Country, Catalonia, and Castile and León.
Significant network outages continued into the night in parts of Andalusia and Galicia, meanwhile, mirroring the slower pace of power restoration in these regions. In some parts of Andalusia, for example, more than half of subscribers remained without service by 6:00 a.m. the following morning, only regaining connectivity once power was restored shortly afterwards. The direct relationship between the timing of power restoration and mobile network recovery highlights that operators are wholly dependent on the grid for service continuity once backup systems are inevitably depleted.
Breadth and Depth of Power Autonomy Drives Differences in Network Resilience Outcomes
The infection curve analogy provides a potent mechanism to visualize and compare network outage profiles and site resilience among Spain’s mobile operators during the blackout. The shape of these outage curves transforms the ‘no service’ data into a story about both network topology and the breadth and depth of power backup deployed across different operators.
In the context of the outage curves, the point where each curve rises from the baseline reflects the level of power autonomy or battery/generator depth. The height of the peak shows how evenly that autonomy is distributed across the site footprint, while the length of the tail can reveal the geographical skew of subscribers within each operator’s network.
On April 28th, Vodafone exhibited the earliest and lowest outage peak in Spain. The probability that a subscriber on its network was left without service peaked at about half that of Orange and notably below other operators. This early yet suppressed peak indicates a thin but widely deployed power autonomy layer in its network, with small cells (featuring short UPS reserves) dropping out quickly, while most large macro sites carrying modest batteries or generators with different configurations likely kept at least one carrier alive for several hours. This approach of spreading shallow reserves across most sites flattened Vodafone’s outage peak but also pulled it forward in time.
While Orange’s outage curve did not reach its peak until several hours after Vodafone, it is indicative of a strategy that concentrated on thick battery or generator backup at key sites but left a broad swath of sites vulnerable to synchronized collapse (as reflected in the sharp no service spikes), the higher peak suggests a larger share of its subscriber base was likely left with no service during the blackout.
Movistar’s outage profile fell between Vodafone and Orange in both the timing and height of its peak during the blackout, but the recovery of its site footprint was significantly more protracted—taking nearly twice as long (almost a day and a half) as the other operators for the proportion of subscribers with no service to drop below 2%. This is likely an artifact of the unique scale of its rural site and subscriber footprint in Spain, where power backup is often more reliant on generators that require manual intervention and are more limited in deployment due to the economic challenges posed by installation in remote areas.
Network Resilience is about more than Power Redundancy
While the Iberian blackout may have been a black swan event, the wider global trend of escalating climate, energy, and security-driven shocks impacting multiple layers of telecom infrastructure—often beyond the direct control of mobile operators—has elevated network resilience from a secondary consideration to a core design principle. This shift is reflected in more robust policy oversight (as exemplified in countries like Norway and Finland) and is likely to be underpinned by fiscal subsidies that seek to support the deployment of more power resilience solutions at mobile sites.
Network resilience is, however, more than keeping the lights on; it also depends on other important factors like having diverse, independent paths to the wider internet. On the day of the blackout, Moroccan operators that funnel most of their international traffic through Spanish landing stations lost those paths when the grid collapse knocked Spanish routers and data centers offline.
Since the subsea capacity and routing of some Moroccan operators was dependent on a highly concentrated upstream Spain-centered corridor, there was limited immediate fail-over and users experienced sharp service degradation, even though power and equipment inside Morocco never went down. By contrast, operators that had additional fibers to France or Italy suffered more minor disruptions, highlighting how geographic and upstream diversity are just as critical to mobile network resilience as local power autonomy.
Cómo Se Comportaron Las Redes Móviles Españolas Durante El Colapso De La Red Ibérica
La variación en la escala y duración de las interrupciones entre los operadores de telefonía móvil durante el evento del 28 de abril refuerza la idea de que una sólida capacidad de recuperación energética es ahora el factor más decisivo para garantizar la continuidad del servicio en las redes móviles.
El colapso de la red ibérica es la demostración más clara de la fragilidad de la infraestructura de las redes móviles precisamente cuando más se necesita. El debate público tras el histórico suceso se ha centrado en comprender y, en última instancia, mitigar las vulnerabilidades sociales derivadas de la dependencia inherente de las redes móviles de un suministro constante de energía de la red, que es cada vez más difícil de garantizar a medida que los incidentes graves se hacen más frecuentes.
Por primera vez, y con el objetivo de fomentar una comprensión más profunda de la interdependencia entre la infraestructura móvil y la eléctrica en Europa, Ookla publica datos detallados que muestran la escala y la distribución geográfica de las interrupciones de la red móvil causadas por el apagón del 28 de abril. Sobre la base de nuestro análisis anterior, esta investigación revela que el impacto del corte de energía en la infraestructura de red móvil varió significativamente entre los operadores, y que los diferentes niveles de penetración de la energía de respaldo son probablemente un factor crítico que determina la gravedad que experimentan los usuarios finales.
Principales Conclusiones:
En el punto álgido del apagón del 28 de abril, más de la mitad de los usuarios de redes móviles de amplias zonas de España se quedaron completamente sin señal móvil. La proporción de usuarios de redes móviles que experimentaron una pérdida completa del servicio (sin poder llamar, enviar mensajes de texto o utilizar datos debido a que los sitios se habían quedado sin señal) pasó de un valor de referencia previo al apagón inferior al 0,5% a más del 50% en amplias zonas de España en el punto álgido del 28 de abril. Esto indica un colapso grave, generalizado e histórico de la red de telefonía móvil que se agravó durante la tarde y la noche del día del apagón a medida que se agotaban progresivamente las reservas de energía disponibles.
El momento y la distribución de los cortes de la red móvil siguieron el patrón de los incidentes de la red eléctrica, lo que evidencia la vulnerabilidad de la infraestructura de telecomunicaciones a las interrupciones, incluso breves, del suministro eléctrico. En los 30 minutos siguientes al fallo de la red (a las 13:00 CET), la proporción de usuarios sin servicio aumentó a medida que se desconectaban las células pequeñas y los emplazamientos con un mínimo de reserva de energía y autonomía de batería. Al cabo de dos horas, alrededor del 12% de los usuarios del operador más afectado carecían de servicio. El aumento de los cortes se ralentizó a continuación, lo que sugiere que los macroemplazamientos restantes, probablemente equipados con bancos de baterías de cuatro a seis horas de autonomía, siguieron funcionando hasta que se agotaron sus reservas, lo que provocó una última y brusca oleada de pérdidas de servicio a última hora de la tarde. El restablecimiento de los servicios móviles siguió el ritmo de la reenergización geográficamente escalonada de la red eléctrica, con interrupciones de la red que se prolongaron hasta bien entrada la noche en partes de Andalucía y Galicia.
Aunque durante el apagón se observaron graves cortes de red en todos los operadores españoles, los usuarios de telefonía móvil de la red de Vodafone tuvieron menos probabilidades de sufrir una pérdida completa del servicio. Entre cuatro y ocho horas después del colapso de la red, el intervalo en el que cada operador alcanzó su peor punto en términos de pérdida de servicio, los abonados de Vodafone tuvieron, de media, menos de la mitad de probabilidades de quedarse sin servicio que los abonados de la red de Orange y notablemente menos probabilidades también que los abonados de Movistar o Yoigo.
La red de telefonía móvil de Marruecos permaneció operativa durante todo el 28 de abril, ya que el suministro de la red nacional no se vio afectado. Sin embargo, la dependencia del país de España para la conectividad internacional en capas más profundas de la red provocó fallos en cascada y una grave degradación del servicio. La proporción de usuarios de telefonía móvil sin servicio en las redes de Orange Maroc y Maroc Telecom se mantuvo el 28 de abril en el mismo nivel que antes del apagón, lo que confirma que no se produjeron interrupciones sostenidas de la red en los emplazamientos de telefonía móvil debido a fallos de la red, a diferencia de lo ocurrido en España y Portugal. Sin embargo, el análisis de los datos de Speedtest Intelligence revela que en Marruecos se siguió produciendo una degradación significativa del rendimiento, con un aumento de más del 20% en el tiempo medio de carga de los sitios web más populares en comparación con el mismo día de la semana anterior. La calidad de la experiencia (QoE) de los abonados de Orange Maroc se vio especialmente afectada durante el apagón, al parecer debido a interrupciones en la conectividad submarina ascendente entre Marruecos y España continental.
La resiliencia eléctrica y las estrategias de gestión de la energía determinan la anatomía de los cortes de red
El colapso de la red ibérica ejemplifica un tipo de prueba de resistencia cada vez más habitual en Europa: la capacidad de la infraestructura de redes móviles para resistir perturbaciones externas graves y prolongadas más allá del operador directo, garantizando la continuidad del servicio, precisamente cuando es más crítico para la seguridad pública y el funcionamiento de la sociedad.
Más allá del apagón del 28 de abril, los recientes sucesos en el Reino Unido e Irlanda, donde las tormentas invernales (especialmente la tormenta Éowyn) causaron grandes daños localizados en las redes de distribución eléctrica, y en Francia, donde el vandalismo en subestaciones provocó breves pero graves apagones en Cannes y Niza, ponen de manifiesto que las interrupciones del suministro eléctrico son una de las principales vulnerabilidades de las redes móviles que dependen totalmente de la red eléctrica. Estas interrupciones se suman a otras vulnerabilidades externas a las que se enfrentan los operadores, como la conectividad de fibra terrestre o submarina, los enlaces ascendentes en la nube y las conexiones igualitarias de terceros.
La resiliencia energética, en particular mediante sistemas de baterías y generadores de reserva en los emplazamientos móviles, ha surgido como una medida proactiva clave para mitigar los cortes de red derivados de las interrupciones de la red eléctrica. En pocas palabras, la energía de reserva desempeña un papel análogo al de las medidas de salud pública en una pandemia: retrasa estratégicamente el inicio de los cortes (ampliando las horas operativas antes de que los sites se queden sin energía) y reduce la gravedad máxima (limitando el número total de sitios afectados simultáneamente).
Por tanto, las inversiones de capital en herramientas de refuerzo de la red, como la redundancia de energía, pueden “aplanar” la curva de impacto del servicio durante un apagón, del mismo modo que las medidas de salud pública aplanaron la curva de infección durante la pandemia de coronavirus. Sin embargo, las interrupciones prolongadas y generalizadas de la red, como la del 28 de abril, demuestran que ninguna medida por sí sola es la panacea y que la autonomía energética a largo plazo no suele ser económicamente viable en una gran proporción de la huella del emplazamiento.
Conscientes de ello, los operadores de telefonía móvil suelen aplicar medidas agresivas de gestión de la energía durante las interrupciones del suministro para maximizar el tiempo de actividad de los emplazamientos y asignar estratégicamente los recursos de red en función de la prioridad de los usuarios. Estas medidas pueden incluir el estrangulamiento de la potencia de transmisión del emplazamiento para reducir las huellas de cobertura, la limitación de la diversidad del espectro para limitar la agregación de portadoras y la capacidad global, y la desactivación temporal de tecnologías más nuevas como 5G y Massive MIMO para ampliar el tiempo de funcionamiento de la batería y el generador.
Las diferencias en el alcance del despliegue de respaldo de energía y el uso estratégico de medidas de conservación de la energía y reducción de la carga conforman la anatomía de los cortes de red. Estas variaciones entre operadores pueden deberse a la diversidad de combustibles elegidos para la energía de reserva (por ejemplo, las baterías pueden ofrecer un potencial de monetización a largo plazo, incluidas las oportunidades de reventa a la red, pero suelen tener tiempos de funcionamiento más cortos), configuraciones de red (como los acuerdos de compartición de RAN y los tipos de emplazamientos, con pequeñas células urbanas densas que se enfrentan a mayores limitaciones físicas para los despliegues de energía de reserva) y características de la base de abonados (por ejemplo, los operadores que atienden a una base de abonados rurales más grande se enfrentan a retos más complejos y costosos para mejorar la resiliencia de la red).
El colapso de la red ibérica se propagó en cascada a través de las redes móviles en España, moviéndose al unísono con las interrupciones del suministro eléctrico
El análisis de los datos de escaneo de señales de fondo de Ookla® revela que el incidente del 28 de abril provocó una tensión sin precedentes en las redes móviles de España, desencadenando un rápido colapso en la densidad de la red. La escala histórica de este colapso redujo gravemente la huella de cobertura de la infraestructura del operador, empujando a una parte significativa de los abonados móviles españoles a un estado de “sin servicio”, incapaces de conectarse a un sitio móvil cercano y, por tanto, temporalmente incapaces de hacer llamadas, enviar mensajes de texto o utilizar datos.
La secuenciación geoespacial de los datos de “sin servicio” calcula la probabilidad media de que un usuario de red móvil se quedara sin señal durante una ventana temporal determinada y sirve como indicador de la proporción global de abonados que se quedan sin acceso a la red. Esta metodología ilustra que el impacto del apagón se propagó rápidamente en cascada por las redes móviles españolas (reflejando fielmente la analogía de la curva de infección descrita anteriormente):
Línea de base (antes del apagón). Los sitios móviles funcionaban con normalidad. La proporción media de abonados sin servicio era inferior al 0,5%, lo que refleja los altísimos niveles de disponibilidad de red típicos de España.
Impacto inicial (inmediatamente después del apagón). La red se colapsó en torno a las 12:33 CET en toda la Península Ibérica, lo que provocó que los rectificadores o los dispositivos de alimentación ininterrumpida (SAI) de las ubicaciones móviles pasaran a utilizar baterías o generadores de reserva (cuando éstos estaban instalados y adecuadamente cargados o alimentados).
A los 30 minutos de la caída de tensión, la proporción de abonados sin servicio había superado con creces la línea de base anterior al incidente en toda España y la curva de cortes entró en una fase de crecimiento descontrolado. Es probable que este impacto inicial limitado de los cortes reflejara la pérdida inmediata de una pequeña parte de la huella total del emplazamiento en zonas en las que no había suministro eléctrico de reserva o su capacidad era muy limitada, normalmente en emplazamientos de células pequeñas en entornos urbanos densos, donde el espacio físico restringe la instalación de baterías o generadores, o en emplazamientos rurales remotos sin redundancia (o con generadores que tardaron más en conectarse).
Cabe destacar que la distribución geográfica equilibrada de los abonados que empezaron a quedarse sin servicio en sus dispositivos (es decir, a perder el acceso a la red) al inicio de los cortes de red sugiere que el momento de los impactos fue en general coherente en toda España, afectando tanto a regiones rurales como urbanas. Sin embargo, la gravedad inicial fue más pronunciada en las zonas situadas a lo largo de las costas este y sur, así como en el interior del país, lo que podría indicar un menor nivel de autonomía eléctrica en los emplazamientos de estas zonas.
Recuperación (2-8 horas después del apagón). A las dos horas del apagón, el operador más afectado ya había visto cómo en torno al 12% de los usuarios se quedaban sin servicio. A partir de ese momento, el ritmo de crecimiento de los cortes disminuyó temporalmente durante varias horas. Este patrón sugiere que el perfil de reserva de energía de los sitios móviles españoles era bimodal o trimodal, con un enfoque escalonado de la capacidad de reserva empleada por los operadores (es decir, la autonomía de los sitios se agrupaba en configuraciones como resiliencia de energía de dos, cuatro o seis horas, con tasas de descarga de la batería determinadas por la capacidad de amperios-hora y la carga del sitio en cada ubicación).
Como resultado de este enfoque escalonado de la autonomía energética y de la variabilidad de la carga del emplazamiento, una parte de los emplazamientos siguieron funcionando, probablemente en un estado reducido con medidas agresivas de conservación de la energía (reflejadas en la marcada degradación del rendimiento observada en los datos de Speedtest Intelligence®), durante toda la tarde y noche del 28 de abril.
El agotamiento final de las grandes reservas de energía, que suelen encontrarse en los principales macro-emplazamientos situados en torres de celosía o monoposte, desencadenó una última oleada de cortes de red entre cuatro y ocho horas después del inicio del apagón. La pérdida de estos emplazamientos “paraguas” de área amplia por la noche provocó que una parte sustancial de los abonados móviles españoles se quedaran sin servicio a las 21:00 CET, con graves cortes de red distribuidos geográficamente por toda España.
Pico (8-10 horas después del apagón). El impacto del apagón en la infraestructura de la red móvil en España alcanzó su punto álgido entre las 21:00 y las 22:00 CET, con más de la mitad de todos los abonados móviles sin servicio en muchas regiones a última hora de la tarde, justo antes de que empezara a volver la electricidad. Para entonces (unas diez horas después de la pérdida inicial de energía), incluso las baterías de reserva más potentes estaban probablemente agotadas, por lo que sólo quedaban operativas las instalaciones alimentadas por generadores móviles o fijos (especialmente en los casos en que el suministro estratégico de combustible podía prolongar el tiempo de actividad).
Recuperación (resto de los días 28 y 29 de abril). El análisis cruzado de la secuencia geoespacial de los datos de ausencia de servicio con las emisiones de luz nocturnas detalladas en las imágenes de satélite de la NASA revela que la recuperación de los servicios móviles y la huella de la red de los emplazamientos en España siguieron de cerca el patrón de reenergización de la red. La proporción de usuarios sin servicio empezó a disminuir a media tarde y a primera hora de la noche en las zonas donde primero se restableció el suministro eléctrico, a medida que el operador de la red realizaba un arranque en negro por fases en todo el país. Esta recuperación temprana de la red fue más evidente en regiones como el País Vasco, Cataluña y Castilla y León.
Mientras tanto, en algunas zonas de Andalucía y Galicia los cortes de red continuaron durante la noche, reflejando el ritmo más lento del restablecimiento del suministro eléctrico en estas regiones. En algunas zonas de Andalucía, por ejemplo, más de la mitad de los abonados seguían sin servicio a las 6 de la mañana del día siguiente, y sólo recuperaron la conectividad cuando se restableció el suministro poco después. La relación directa entre el momento del restablecimiento del suministro eléctrico y la recuperación de la red móvil pone de manifiesto que los operadores dependen totalmente de la red para la continuidad del servicio una vez que los sistemas de reserva se agotan inevitablemente.
La amplitud y la profundidad de la autonomía eléctrica determinan las diferencias en los resultados de la recuperación de la red
La analogía de la curva de infección proporciona un potente mecanismo para visualizar y comparar los perfiles de interrupción de la red y la capacidad de recuperación de los emplazamientos entre los operadores móviles españoles durante el apagón. La forma de estas curvas de cortes transforma los datos de “ausencia de servicio” en una historia sobre la topología de la red y la amplitud y profundidad de la reserva de energía desplegada por los distintos operadores.
En el contexto de las curvas de interrupciones, el punto en el que cada curva se eleva desde la línea de base refleja el nivel de autonomía eléctrica o la profundidad de la batería/generador. La altura del pico muestra hasta qué punto la autonomía se distribuye uniformemente a través de la huella del sitio, mientras que la longitud de la cola puede revelar el sesgo geográfico de los abonados dentro de la red de cada operador.
El 28 de abril, Vodafone exhibió el pico de cortes más temprano y más bajo de España. La probabilidad de que un abonado de su red se quedara sin servicio alcanzó un máximo de aproximadamente la mitad que el de Orange y notablemente por debajo de otros operadores. Este pico, temprano pero suprimido, indica una capa de autonomía energética delgada pero ampliamente desplegada en su red, con células pequeñas (que cuentan con reservas cortas de SAI) que se quedan sin servicio rápidamente, mientras que la mayoría de los grandes macroemplazamientos que llevan baterías modestas o generadores con diferentes configuraciones probablemente mantuvieron vivo al menos a un operador durante varias horas. Este planteamiento de repartir reservas poco profundas entre la mayoría de los emplazamientos aplanó el pico de cortes de Vodafone, pero también lo adelantó en el tiempo.
Aunque la curva de cortes de Orange no alcanzó su pico hasta varias horas después que la de Vodafone, es indicativa de una estrategia que se concentró en una reserva de baterías o generadores de gran capacidad en los emplazamientos clave, pero que dejó una amplia franja de emplazamientos vulnerables al colapso sincronizado (como reflejan los fuertes picos sin servicio), el pico más alto sugiere que una mayor parte de su base de abonados probablemente se quedó sin servicio durante el apagón.
El perfil de interrupción de Movistar se situó entre el de Vodafone y Orange tanto en el momento como en la altura de su pico durante el apagón, pero la recuperación de su huella de sitios fue significativamente más prolongada, tardando casi el doble de tiempo (casi un día y medio) que los otros operadores para que la proporción de abonados sin servicio cayera por debajo del 2%. Es probable que esto se deba a la escala única de su presencia rural y de abonados en España, donde la energía de reserva suele depender más de generadores que requieren intervención manual y cuyo despliegue es más limitado debido a las dificultades económicas que plantea la instalación en zonas remotas.
La resiliencia de la red va más allá de la redundancia energética
Si bien el apagón en la Península Ibérica pudo haber sido un evento inesperado, la creciente tendencia global de crisis climáticas, energéticas y de seguridad que impactan múltiples capas de la infraestructura de telecomunicaciones —a menudo fuera del control directo de los operadores móviles— ha elevado la resiliencia de la red de una consideración secundaria a un principio de diseño fundamental. Este cambio se refleja en una supervisión política más exhaustiva (como se ejemplifica en países como Noruega y Finlandia) y es probable que se sustente en subsidios fiscales que buscan apoyar el despliegue de más soluciones de resiliencia energética en las estaciones móviles.
La resiliencia de la red, sin embargo, va más allá de mantener las luces encendidas; también depende de otros factores importantes, como contar con rutas diversas e independientes hacia una internet más amplia. El día del apagón, los operadores marroquíes que canalizan la mayor parte de su tráfico internacional a través de estaciones de aterrizaje españolas perdieron esas rutas cuando el colapso de la red dejó fuera de servicio a los routers y centros de datos españoles.
Dado que la capacidad submarina y el enrutamiento de algunos operadores marroquíes dependían de un corredor de aguas arriba altamente concentrado y centrado en España, la conmutación por error inmediata fue limitada y los usuarios experimentaron una fuerte degradación del servicio, a pesar de que el suministro eléctrico y los equipos dentro de Marruecos nunca se interrumpieron. Por el contrario, los operadores que contaban con fibra adicional con Francia o Italia sufrieron interrupciones menores, lo que pone de manifiesto cómo la diversidad geográfica y de aguas arriba es tan crucial para la resiliencia de la red móvil como la disponibilidad de energía local.
Ookla retains ownership of this article including all of the intellectual property rights, data, content graphs and analysis. This article may not be quoted, reproduced, distributed or published for any commercial purpose without prior consent. Members of the press and others using the findings in this article for non-commercial purposes are welcome to publicly share and link to report information with attribution to Ookla.
Luke Kehoe leads Ookla’s research and thought leadership efforts in Europe.
An electronic engineering alumnus of University College Dublin, Luke has extensive experience collaborating with mobile operators, telecoms vendors, and government agencies in research and advisory roles across Europe. He has contributed to internationally recognised thought leadership publications in areas such as 5G, IoT, open RAN, and edge computing, working with prestigious organisations like the Telecom Infra Project and the World Economic Forum.
Indoor connectivity challenges have intensified as modern insulation materials, the shift to mid-band spectrum, and the sunset of 3G networks prevent outdoor mobile sites from reliably penetrating buildings
As much as 80% of all mobile data usage originates from indoor environments like homes, offices and shops. However, mobile networks were initially designed with an ‘outside-in’ approach—relying on outdoor towers to deliver coverage, with the expectation that the signal would reach indoors without being specifically optimised to do so. This strategy helped minimise deployment costs and was based on the assumption that indoor connectivity could be provided by low-band spectrum layered over the macro mobile network, with higher data rate demands met by home broadband and public Wi-Fi networks indoors.
Consumers have come to rely on mobile data to serve their indoor browsing needs and expect performance parity as they move around from home, work, the shops, and everywhere in between. Even where Wi-Fi and related features like VoWiFi are available and sufficiently fast, in-building mobile coverage remains critical for last resort access to basic telephony features like calling and texting to ensure reliable access to emergency service networks. Indeed, in many advanced European markets, operators and regulators prioritise routing 112 emergency calls over mobile networks using VoLTE rather than Wi-Fi, as VoLTE offers greater reliability and quality of service through dedicated voice packet routing on mobile networks.
But if indoor connectivity is so important, why is it still so lacklustre? While there is no one easy answer, there are a few clear contributing factors.
More mid-band spectrum in 5G networks introduces new propagation challenges
One of the biggest barriers to good indoor connectivity lies in how networks are designed, and this challenge is becoming more common with the deployment of 5G. The trend towards higher frequency spectrum for 5G (e.g. 3.5 GHz mid-band) limits the ability of the existing mobile network site grid to provide high-speed mobile coverage deep indoors. This is due to the more constrained propagation characteristics of this spectrum. Simply put, the signals that mid-band 5G networks rely on struggle to penetrate the materials in their path when the user is indoors.
Lower frequency signals do not face this problem to the same extent, but their utility has become more limited over time. While the lower frequency spectrum (e.g. 800/900 MHz with 3G/4G and 700 MHz more recently with 5G) traditionally used to provide in-building mobile coverage previously sufficed, the significant increase in the density of devices and the intensity of their data traffic demands mean these frequencies alone are unable to support the higher performance attributes often expected with 5G, particularly in dense urban settings.
Because of this, the traditional approach of outside-in network design, where signals are transmitted from the macro coverage layer of a lattice or monopole-based high site into a cluster of buildings, is no longer fit for purpose in the absence of investment in network densification if demands for reliably fast connectivity indoors are to be met.
Modern insulation materials turn buildings into Faraday cages
Network design is not the only contributing factor to the profile of signal propagation. While it is true that the signals typically used for 5G networks struggle to travel through buildings, some materials present a bigger challenge than others.
The use of modern insulation materials in new-build and retrofitted developments is posing a significant challenge for mobile operators. Take low-E glass, for example – a type of energy-efficient glass with a microscopic coating designed to reduce energy consumption, which is becoming a commonplace alternative to double glazing. Low-E glass has a significant negative impact on radio signal propagation, and with its growing use in retail and office buildings, the indoor connectivity problem is set to worsen, especially with the use of higher frequency bands
As these kinds of construction materials – those that significantly increase signal attenuation and effectively turn buildings into Faraday cages – become more widely used, network design and building design must go hand-in-hand. Otherwise, the ability of 5G signals to penetrate newer buildings will continue to be diminished.
Technology sunsets require deep network modernization to replicate legacy coverage footprints
The sunset of legacy network technologies like 2G (in markets such as Switzerland and the US) and 3G (in most developed markets) has introduced further challenges as operators seek to preserve indoor coverage levels while upgrading equipment and repurposing frequencies.The process of improving network performance and optimising long-term operating costs with technology sunsets is not as simple as removing and replacing outdated equipment. Operators need to ensure legacy end user devices are upgraded to take advantage of 4G and 5G networks and that older mobile sites are refreshed with modern radio equipment to ensure there is full continuity in coverage levels.
Time Without Service Rose Across All Polish Operators in 2024 as the 3G Sunset Advanced
Speedtest Intelligence® | FY 2023 – 2024
Analysis of Speedtest Intelligence data has revealed a concerning trend of increased time spent on 2G networks or with no service at all in several advanced markets where operators have been slower to repurpose spectrum employed by legacy technologies upon sunsetting 3G. This has manifested in increased reports of dropped calls and other mobile connectivity issues, particularly in areas where decommissioned 3G coverage has yet to be fully replaced by 4G or 5G networks.
Policy goals and incentives place emphasis on outdoor coverage, treating indoor access as incidental
Governments and regulators around the world have historically focused headline policy goals on achieving outdoor population coverage targets. This model has overlooked the importance of indoor mobile coverage, contributing to poor outcomes throughout in-building environments and a lack of public data on the extent of indoor coverage gaps. Some countries, like Ireland and Germany, have made progress by mandating minimum coverage levels at buildings and infrastructure of national importance as part of spectrum licence conditions. In the Irish context, for example, this includes a requirement to provide a minimum 30 Mbps service across key infrastructure sites like train stations and hospitals, as well as community hubs and tourist locations.
These types of progressive policies, as well as those being adopted by city governments to increase building access for mobile sites through amendments to planning and zoning conditions on future renewals and large-scale commercial and residential developments, can play a positive role in stimulating better indoor coverage outcomes by re-aligning deployment incentives and removing obstacles.
New deployment models, richer data insights, and greater policy oversight can drive better indoor outcomes
While consumers expect consistently high-performing in-building mobile performance, the path to get there is not a simple one. There is no one-stop solution to the indoor connectivity problem.
That said, the neutral host model is emerging as a key solution to improve in-building mobile outcomes, providing multi-operator access to promote fair competition and share deployment costs, typically based on small cell solutions like the Ericsson Radio Dot. Freshwave (UK) and Proptivity (Sweden) are early examples of neutral host specialists leading the charge in this space.
While the scaling up of small cell deployments at the street and building level, enabled by the neutral host model, is key to improving indoor performance, there are other factors at play. Operators must prioritise repurposing the spectrum in the wake of 3G sunsetting, and building developers and the planning system should take better account of the accommodations needed to host radio equipment. But if indoor connectivity is truly to see a material improvement, these changes should be underpinned by progressive regulatory policies that measure indoor coverage levels and provide better incentives to improve in-building mobile outcomes and remove barriers to deployment.
Ookla retains ownership of this article including all of the intellectual property rights, data, content graphs and analysis. This article may not be quoted, reproduced, distributed or published for any commercial purpose without prior consent. Members of the press and others using the findings in this article for non-commercial purposes are welcome to publicly share and link to report information with attribution to Ookla.
Luke Kehoe leads Ookla’s research and thought leadership efforts in Europe.
An electronic engineering alumnus of University College Dublin, Luke has extensive experience collaborating with mobile operators, telecoms vendors, and government agencies in research and advisory roles across Europe. He has contributed to internationally recognised thought leadership publications in areas such as 5G, IoT, open RAN, and edge computing, working with prestigious organisations like the Telecom Infra Project and the World Economic Forum.