| May 27, 2026

Romania’s Mobile Reset: How Market Consolidation is Shaping 5G Rollout

Romania ranks poorly in mobile performance compared to its Central and Eastern European peers.

Romania’s mobile market has entered a more consequential phase of its 5G cycle. The country has moved beyond the initial spectrum allocation stage, 5G adoption is increasing, and the market structure changed after Vodafone and DIGI divided Telekom Romania Mobile between them in 2025. Vodafone took the postpaid and business customer base and much of the network infrastructure, while DIGI took the prepaid business and selected spectrum and tower assets.

Q1 2026 is the first full quarter of the three-mobile operator structure in Romania. Analysis of Speedtest Intelligence data reveals that the country still has a material gap to close, with Romania ranking near the bottom of Central and Eastern European (CEE) peers for overall mobile performance and 5G coverage, and with quality of experience (QoE) results weaker than its fixed broadband reputation might lead policymakers or enterprise users to expect.

It is still early post-consolidation, but the first signals are emerging on whether Orange, Vodafone, and DIGI can turn different spectrum portfolios, site grids, refarming strategies, and integration paths into better mobile service.

Key Takeaways:

  • Romania’s mobile performance gap is driven as much by 5G reach as by headline speed. In Q1 2026, 5G Availability stood at 39%, placing Romania toward the lower end among Central and Eastern European peers. That limited 5G reach helps explain why Romania’s median mobile download speed was also the lowest in the group, at 78.43 Mbps.
  • Orange has the broadest 5G footprint among Romania’s operators, but this has not automatically translated into application latency leadership. Orange led on 5G Availability (57%) and overall median mobile download speed (97.70 Mbps) in Q1 2026, but it lagged on real-world latency outcomes measured to cloud infrastructure endpoints like AWS and Azure.
  • Vodafone’s 5G footprint is narrower, but its 5G throughput result points to substantial capacity where 5G is deployed. Vodafone recorded median 5G download speeds of 180.57 Mbps in Q1 2026 (doubling year on year), with 36% 5G Availability, making network integration and spectrum refarming after the Telekom Romania Mobile deal central to further footprint expansion.
  • DIGI’s results show why quality of experience does not always map neatly to download speed. DIGI recorded a lower 5G median download speed of 126.51 Mbps, but it had the highest Romanian operator 5G upload result at 33.03 Mbps and the lowest 5G cloud latency at 79 ms, potentially reflecting the merits of a dense fiber network, deeper peering or better optimized traffic routing and interconnection.
  • Romania’s 5G gap is not only a question of rollout. It also reflects a thinner spectrum base in operators’ commercial portfolios. Only 30 MHz of national 700 MHz spectrum was assigned in the 2022 5G auction, around half the CEE norm of 60 MHz, after Telekom Romania Mobile chose not to apply. Romania also has the lowest assigned C-band depth below 3.7 GHz in the peer group, at 310 MHz.

Telekom Romania split creates a new market dynamic

Romania consolidated from four mobile operators to three on 1 October 2025, when Vodafone Romania completed a €30 million (US$34 million) purchase of Telekom Romania Mobile Communications from OTE and DIGI Romania paid €40 million (US$45 million) for a parallel carve-out from the same target. Vodafone took the legal entity, the postpaid base, and most of the spectrum.

DIGI took the prepaid book, residual towers, and spectrum blocks at 900, 1800, 2100, and 2600 MHz. The Romanian Competition Council cleared the deal on 28 July 2025, conditional on continued MVNO wholesale access, Orange co-location, DIGI road-coverage investment, and a four-year compliance trustee. ANCOM issued the modified licences on 8 October.

The deal closed three years of strategic re-orientation by parent OTE, which had retained the mobile arm after selling fixed-line Telekom Romania to Orange in 2021. That earlier sale produced a single convergent Orange footprint, legally merged on 1 June 2024. OTE ran Telekom Romania Mobile as a sub-scale fourth challenger with roughly 12% of active SIMs by end-2024, and its non-participation in the November 2022 5G auction left Romania with 30 MHz at 700 MHz, half the CEE norm.

The transaction did not come out of nowhere. Vodafone had already signaled that Romania was a market where consolidation could reshape the business. In its FY25 results, Vodafone Group disclosed a combined €4.5 billion goodwill impairment across Germany and Romania, with Romania specifically identified as a market offering “the opportunity to drive further consolidation.” The Telekom Romania Mobile transaction then became the mechanism for that consolidation, with Vodafone strengthening its mobile position and DIGI acquiring assets that helped address a long-standing weakness in its mobile spectrum base.

By Vodafone’s H1 FY26 results on 11 November 2025, 250,000 Telekom Romania customers had been migrated. The Telekom brand persists pending full legal merger into Vodafone after 1 July 2026. For our Q1 2026 analysis, Telekom Romania is treated as transitional rather than a continuing peer.

5G speeds are lagging but year-over-year results show promise

Romania’s overall median mobile download speed of 78.43 Mbps in Q1 2026 placed it last among the eight CEE countries examined in our analysis . The gap to the leader was large. Bulgaria delivered 196.35 Mbps (among the highest in Europe and across advanced liberal democracies), Serbia 134.92 Mbps, and Croatia 126.02 Mbps. Hungary, with its similar three-operator structure and rural geography, reached 83.24 Mbps. Romania’s 5G median of 144.55 Mbps ranked second-last, ahead only of Czechia at 113.67 Mbps, in a cohort where Serbia delivered 340.92 Mbps and Bulgaria 303.41 Mbps on 5G.

Analysis of median multi-server latency outcomes tells a similar story. Romania at 41 ms had the highest median latency in the CEE cohort, with the next-worst result (Poland, 39 ms) measurably ahead. On 5G, Romania at 39 ms also sat at the high-latency end of the cohort.

Our quality of experience (QoE) data reinforces this conclusion. Romania ranked last among the peer cohort for overall web page load times and second-last for 5G web page load times to popular websites. It also ranked near the bottom for 5G cloud latency (measuring to major hyperscaler endpoints like AWS, Azure, Google Cloud and Oracle) and for the share of mobile users who experience Full HD video resolution (1080p or higher) on 5G.

These results suggest that the country’s 5G gap is not just a radio access issue but also an application path issue shaped by deeper factors like transport, routing, peering, CDN placement, core architecture, and how much of the user journey actually stays on a well-performing 5G layer.

Vodafone's 5G Speeds Doubled After Absorbing Telekom Romania's Spectrum
Speedtest Intelligence® | Q1 2025–Q1 2026

Two contrasting patterns are, however, worth surfacing. The first is year-on-year improvement. Romanian median mobile download speeds rose 12% from 69.79 Mbps in Q1 2025 to 78.43 Mbps in Q1 2026, with 5G median speeds rising 9% over the same period from 132.91 Mbps to 144.55 Mbps. These improvements track the operator-level effects of consolidation, where Vodafone absorbed spectrum and density, and DIGI extended its coverage layer.

Romania Sits Near the Bottom of CEE for 5G Availability
Speedtest Intelligence® | Q1 2026

The second notable trend is observable in floor performance. Romania’s 10th percentile 5G download speed of 28.71 Mbps ranks third in the cohort, behind only Serbia and Bulgaria. In effect, this means that when users connect to 5G, the experience is competitive. The weakness is what happens between 5G sessions, where users still spend a large share of time on 4G layers that lack the dense mid-band overlay seen in CEE leaders like Bulgaria.

Spectrum access continues to trail regional peers

Romania’s total assigned mobile spectrum below 6 GHz amounts to 964 MHz, near the bottom of the CEE cohort. The structural gaps uniquely span both the coverage layer and the mid-band overlay. Sub-1 GHz holdings combined across 700, 800, and 900 MHz come to 160 MHz, second-last in the cohort and only ahead of Serbia.

The country’s 700 MHz allocation (ideal for broad 5G coverage), in particular, is uniquely thin. Romania assigned only 30 MHz at 700 MHz in the 2022 multi-band auction that closed on 15 November 2022, vs. the 60 MHz typical of CEE peers. The reason is documented in ANCOM’s pre-auction notice. Pre-consolidation Telekom Romania Mobile did not submit an application. Only Orange, RCS&RDS (DIGI), and Vodafone bid, and one of the three paired 2×10 MHz blocks went unsold.

C-band mid-band depth is similarly constrained. Romania’s total of 310 MHz across 3.4 to 3.8 GHz is the lowest in the CEE-8 cohort below 3.7 GHz. Czechia, Poland, Greece, Hungary, Serbia, and Slovakia all hold between 390 and 480 MHz at 3500 MHz. Orange Romania holds 160 MHz, Vodafone Romania 100 MHz, and DIGI 50 MHz of that 310 MHz total (exposing huge asymmetries among operators too). The 2022 auction itself raised €432.6 million (US$489 million) against an ANCOM-expected €692 million at reserve prices, a 37% shortfall that signalled operator caution about locking in long-duration spectrum at the offered prices given Romania’s comparatively weak monetization base (low ARPU) and potential pre-consolidation uncertainty.

Coverage obligations require winners to cover at least 70% of the national population, all main highways, modernized railways, international airports, and 240 specifically named under-served settlements, with milestone deadlines through 2030. To that end, overall 5G Availability in Romania reached 39% in Q1 2026 in our data (up four percentage points year on year), below the CEE average but notably ahead of or comparable with mid-band centric 5G performance leaders Bulgaria (36%) and Hungary (41%).

The spectrum consolidation effects of the recent restructures in the market are likely to materially improve DIGI’s ability to compete on 5G coverage and Vodafone’s to compete on capacity. The October 2025 ANCOM license modifications reshaped DIGI’s spectrum portfolio. DIGI gained 2×5 MHz at 900 MHz, 2×15 MHz at 1800 MHz, 2×10 MHz at 2100 MHz, and 2×10 MHz FDD at 2600 MHz from the carve-out, expanding its sub-1 GHz position from 20 MHz to 30 MHz and adding 1800 MHz capacity it did not previously hold. Vodafone, meanwhile, consolidated the residual Telekom holdings, ending Q1 2026 as the operator with the deepest combined sub-1 GHz and 1800 MHz portfolio.

Low ARPU constrains ability to invest in Romania

Romania’s mobile market generates less revenue per connection than any other among the eight-country CEE cohort we examined, reflecting a legacy of fierce price-based competition. GSMA Intelligence reports ARPU per mobile connection at €6.45 (US$7.51) for Q1 2026. The next-lowest is Poland, which sits at €8.65 (US$10.08) per subscriber. Hungary leads at €13.84 (US$16.12), more than twice Romania’s level.

While capital intensity (share of revenues reinvested) appears healthy in relative terms at 18% of mobile revenue in GSMA data, above the CEE median and ahead of Czechia (11%) and Poland (10%), this masks the absolute revenue available to reinvest in the first instance. With per-connection ARPU under €7 (US$8.15), the capex envelope per population point is materially smaller than in CEE peers.

Romania's Mobile ARPU Is Less Than Half the CEE Leader
GSMA Intelligence | Q1 2026

The country’s striking fixed-mobile divergence amplifies this picture. Romania ranks among EU (and global) leaders on fixed broadband, with one of the largest shares of high-speed household connections and third in the EU for gigabit connections based on the European Commission’s DESI data.

DIGI alone covers close to 100% of urban households with FTTH and holds about 70% of fixed connections. Orange reports more than 26,000 km of national fiber backbone and around 85,000 km of metro fiber, with 412,000 FTTH connections modernized during 2025.

In other words, fixed outcomes are excellent, but Romania’s mobile networks are held back by a combination of unfavorable factors coming together. These include the unusually compressed ARPU base noted here, the thin coverage-layer spectrum portfolio mentioned earlier, and a market structure that until October 2025 included a sub-scale fourth operator (Telekom) that could not absorb the investment required to keep pace.

Orange leads on network breadth and overall performance

Orange starts from the broadest Romanian operator position on 5G reach, likely helped by the largest 700 MHz allocation among the three operators (20 MHz, against Vodafone’s 10 MHz and zero at DIGI). In Q1 2026, Orange recorded 5G Availability of 57% (compared to 36% on Vodafone and 25% on DIGI). Its strong overall spectrum position also conferred an advantage in median download performance, leading with 97.70 Mbps across all technologies (as much as 39% and 45% higher than Vodafone and DIGI respectively).

These results are consistent with the operator’s public strategy. Orange Romania said in its 2025 results that 5G and 5G+ services were available in 88 cities, with consolidated turnover reaching €1.46 billion (US$ 1.7 billion), up 3%. Last year, it launched what it described as the country’s first public 5G standalone deployment in the Danube Delta, supported by a 5G core in Bucharest and a local edge node (an EU Connecting Europe Facility project covering 14 of 23 target localities).

Orange Leads on 5G Reach, DIGI Leads 4G Coverage
Speedtest Intelligence® | Q1 2026

The company said its 2026 investment budget would exceed €200 million (US$ 234 million). Its enviable spectrum position (including in mid-band depth and with a 1500 MHz supplemental downlink advantage) and its early 3G sunset therefore gives it several levers to use that capital to continue to compete. However, the caveat is latency and quality of experience, which is not simply shaped by spectrum alone but also other factors related to peering and interconnection.

Orange ranked behind DIGI but ahead of Vodafone on several responsiveness-related metrics, including on median multi-server latency, web page load times and latency to cloud infrastructure endpoints, with DIGI pulling further ahead on 5G.

Vodafone combines high 5G speeds with integration pressure

Vodafone Romania led on median 5G download speed at 180.57 Mbps, double its Q1 2025 result of 90.57 Mbps and surpassing Orange for the first time to become market leader. The doubling is consistent with absorbing Telekom Romania’s spectrum across 800, 900, 1800, 2100, and 2600 MHz, and with the operator’s various integration commitments.

Vodafone signed a six-year 5G RAN partnership with Ericsson on 12 February 2024 covering 5G RAN, Standalone transition, and Massive MIMO. A week later, Vodafone launched commercial Open RAN across 20 Romanian cities with Samsung vRAN, Wind River, and Dell PowerEdge, one of the larger commercial Open RAN footprints in Europe (supporting a more flexible vendor and radio strategy over time in theory, but also potentially introducing additional complexity). CEO Nedim Baytorun has committed €150 million (US$ 170 million) in incremental network investment tied to the integration.

Despite the newfound lead on 5G speed post-consolidation, the operator continues to lag its peers on overall QoE outcomes (although comparable with or ahead of Orange on a small number of measures), suggesting less mature or optimized peering, routing or interconnection depth in Romania to the most popular content and cloud infrastructure. In Q1 2026, for example, median web page load times on Vodafone’s network were as much as 7% slower than on peers.

DIGI’s challenger model and fiber depth produces favorable latency signals

DIGI’s Q1 2026 results reveal a different network thesis, aligning with its smaller mobile scale and narrower spectrum position. While it continues to trail peers on median 5G download speeds (principally due to more limited mid-band assets), it leads on 5G Consistency (the share of Speedtest samples with minimum performance of at least 25 Mbps down and 3 Mbps up) and 10th percentile 5G speeds, suggesting users on its network are less likely to end up with very weak tail outcomes on 5G.

Despite featuring a smaller 5G footprint (held back by the lack of 700 MHz), DIGI led on 4G Availability in Q1 2026 (with ~98% compared to a close ~96% on Orange and ~94% on Vodafone), and exhibited the lowest share of users spending most of their time stuck on legacy 2G networks (a likely merit of its broader 4G coverage layer observed in our data).

It is notable that the operator has pursued an aggressive spectrum refarming strategy, likely stimulated by its leaner portfolio requiring prudent management, being the first to sunset 3G and reuse its 2100 MHz (2×15 MHz) spectrum fully for 4G and 5G. On 5G coverage, DIGI is targeting a ramp from 30% population coverage at the end of last year to 70% by end-2028 and 99.8% by end-2030, underpinned by about 4,500 mobile sites.

The most interesting part of DIGI’s performance profile, however, is latency and QoE. DIGI recorded the lowest Romanian operator median multi-server latency (as much as 10% lower than peers) across all technologies and on 5G. In real-world testing of popular hyperscaler and website infrastructure endpoints, it also recorded the lowest cloud latency (as much as 14% faster) and shortest web page load times on 5G.

The QoE results suggest a relatively better experience for consumers in applications requiring responsiveness like video calling and gaming on DIGI’s network. That does not mean DIGI has the largest radio capacity position overall (evidenced by lower median and 90th percentile speeds), and highlights how application experience depends on more than download speed.

DIGI’s deep fixed network assets, backhaul economics, routing, and traffic management may be helping parts of the application experience even where its 5G download speed trails Orange and Vodafone.

Romania’s weaker mobile performance does not appear to be explained by a lack of local internet exchange, peering or content delivery infrastructure. Within Romania, Bucharest has a competitive peering and CDN ecosystem that, in theory, should not be a major bottleneck for operators.

The InterLAN internet exchange operates 125 peers across 5.4 Tbps of capacity, while RoNIX adds another 1.4 Tbps. Major content and cloud providers, including Google, Cloudflare, Meta, Microsoft, Akamai and Netflix, are also present in Bucharest. This points to operator-level differences in radio access, transport, core architecture, peering policy or traffic management as more likely drivers of QoE variation.

Coverage obligations set the baseline, but operator investment will determine performance

The next 12 to 18 months will test how far the new market structure can move performance in Romania. The Vodafone-Telekom legal merger is scheduled to complete after 1 July 2026, at which point the combined mid-band network should continue to extend the year-on-year 5G gains already visible in Q1 2026.

The timing of the first 5G SA launch and deployment at scale is the larger unknown. Orange’s Danube Delta cluster and Vodafone’s Politehnica private network are proofs of concept, not commercial services, and no Romanian operator has yet commercialized features like Voice over New Radio (VoNR), network slicing, or similar at national scale. Vodafone Group’s five-year programmable networks partnership with Ericsson, signed in October 2025 and naming Romania as a primary market, is the most concrete commitment to closing that gap.

Policy will set the floor moving forward, since ANCOM’s 2022 auction conditions tied spectrum rights to public coverage outcomes. Established 700 MHz winners must cover 95% of assigned under-served settlements, reach areas inhabited by 70% of the population, and meet staged transport-corridor milestones across highways and modernized rail. C-band winners, meanwhile, carry obligations at international airports, including at least 100 Mbps downlink and an 85% indoor reception probability, alongside base-station deployment targets.

Backed by ANCOM’s signal and service verification methodology, those terms should keep 5G from concentrating only in the most commercially attractive urban cores. DIGI’s aforementioned commitment to 99.8% population coverage by end-2030, on a base of about 4,500 antennas, runs in the same direction, reinforced by the road-coverage obligations the Competition Council attached to its Telekom Romania spectrum carve-out.

Economics will continue to shape how far operators go beyond minimum rollout obligations. Romania has some public support for connectivity, but not a broad national subsidy for mobile RAN expansion. The European Commission’s Romania connectivity profile cites €94 million of Recovery and Resilience Plan support for white and grey areas, plus CEF Digital funding for targeted projects such as 5G Connect Danube Delta.

In practice, the main performance lift will still come from operator capital allocation, spectrum deployment, site access, and core (i.e. with 5G SA) evolution. The open question for regulators is whether consolidation actually delivers the investment lift Vodafone’s commentary suggests, or whether the ARPU compression behind its €4.5 billion (US$5.09 billion) impairment continues to bite.

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.

| May 25, 2026

In Europe, Starlink Finds Its Sweet Spot

Speedtest data across 27 European markets shows LEO broadband gaining ground where fiber, cable, and fixed wireless still leave gaps.

Europe’s broadband story can look close to complete from a national coverage chart. Fiber networks are expanding, gigabit targets are embedded in policy, and many households now have much faster fixed broadband than they did a few years ago. The harder question is what happens at the edge of that progress.

Starlink’s role as a low-Earth orbit broadband provider is clearest in those remaining gaps. Across Europe, the service is most visible where fixed networks are delayed, weaker, difficult to extend, or needed as backup. That includes islands, farms, mountain communities, seasonal homes, remote businesses, and rural premises still waiting for planned upgrades.

Key Takeaways

  • Starlink got faster in most European markets. Starlink’s median download speed across all 27 countries rose from 114.05 Mbps in Q1 2025 to 165.71 Mbps in Q1 2026, up 45%. Download speed improved in 26 of the 27 markets.
  • Usage is most visible where fixed broadband leaves more gaps. Bulgaria had the highest Starlink Speedtest sample share at 8%, followed by Greece and Croatia at 6% each, Ireland and Latvia at 4% each.
  • Terrestrial fixed broadband still has major advantages. Starlink was faster than the average fixed network on median download speed in 11 of 27 markets, but fixed networks had lower latency in every market and stronger upload performance in 26.
  • Satellite broadband is becoming a broader European category. While Starlink is the most visible direct-to-consumer low Earth orbit (LEO) provider, Eutelsat OneWeb, Amazon Leo, and Europe’s IRIS² program point to a wider satellite connectivity market starting to emerge.

Starlink speeds improved across most of Europe

Starlink’s performance backdrop is much stronger than it was just a year ago. SpaceX’s 2025 Progress Report said it connected more than 4.6 million new active customers during 2025 and operated more than 9,000 active satellites by year-end. This scale is showing up in our European consumer-initiated performance data, with investments in constellation and ground station expansion driving improved real-world outcomes.

Starlink download speeds improved across most measured European markets
Speedtest Intelligence® | Q1 2025 to Q1 2026

Across the 27 measured markets combined, Starlink’s median speed in Q1 2026 was 165.71 Mbps on download, 24.10 Mbps on upload, and 49 ms on multi-server latency. Latvia led on median download speed at 232.51 Mbps, followed by Greece at 196.31 Mbps and Croatia at 188.02 Mbps. Cyprus, Poland, and Latvia posted the largest year-on-year download gains, rising by 159.53 Mbps, 134.31 Mbps, and 104.38 Mbps respectively.

Bulgaria was the main exception. It had the highest Starlink sample share in Q1 2026, but the weakest Starlink median download speed at 61.06 Mbps. It was also the only measured market where Starlink download speed was lower than a year earlier, down 5%. This combination suggests heavy or concentrated demand can still strain available satellite capacity in some locations.

Hard-to-reach homes continue to drive satellite adoption

Our data shows that the pattern of satellite adoption is more nuanced than a rural map. Starlink uptake is strongest where fixed broadband options are weaker, delayed, or uneven.

Greece is the clearest example of Starlink filling a gap while terrestrial upgrades continue. Starlink recorded a median download speed of 196.31 Mbps in Q1 2026, compared with the national median of 94.29 Mbps across all fixed networks. Greece also had the second-highest Starlink sample share at 6%. This is not surprising in a country dominated by islands, mountains and dispersed settlements, which has made universal high-speed fixed coverage harder, slower, and more expensive to deliver, even as the country’s National Broadband Plan targets at least 100 Mbps service, upgradeable to gigabit, for all buildings.

Starlink sample share is highest in Bulgaria, Greece, and Croatia
Speedtest Intelligence® | Q1 2026, top 15 European markets

Latvia showed a similar gap from a different starting point. Starlink reached a median download speed of 232.51 Mbps there in the same period, 85.64 Mbps faster than the national median. Latvia’s digital connectivity strategy targets equal, high-quality connectivity across the country, while public support focuses on areas where operators lack the economic case to deploy.

Ireland and the UK, by contrast, illustrate how Starlink can sit alongside fiber-led policy. Ireland’s National Broadband Plan — one of the largest state investments in rural broadband anywhere in the developed world — covers more than 560,000 premises, 1.1 million people, over 65,000 farms, 44,000 non-farm businesses, and 679 schools. Despite this, the country ranked fourth for Starlink adoption in Europe in Q1 2026, likely reflecting usage in the areas still waiting for fiber and as a resilient backup option to mitigate against what have become regular fiber line faults during recent winter storms.

The UK, meanwhile, has moved toward a more explicit hybrid model through Project Gigabit, government LEO trials, and a BT Group Starlink agreement for the UK’s hardest-to-reach places. It ranked tenth in Europe for Starlink sample share in the latest quarter, and featured among the best responsiveness outcomes on the service with median multi-server latency of 37 ms, likely a byproduct of the dense ground station and traffic breakout ecosystem there.

Fast, price-competitive fixed networks still limit Starlink’s role in Europe

Starlink’s best European results do not make it a broad replacement for strong fixed broadband. In markets with mature fiber, cable, or high-speed fixed broadband, its footprint is much smaller.

Seven measured markets had Starlink sample share below 1% in Q1 2026: Denmark, Malta, Finland, Romania, Luxembourg, the Netherlands, and Slovenia. In that below-1% group, Starlink trailed the national median download speed by 81.82 Mbps on average.

Starlink is relatively faster where fiber penetration is lower
Speedtest Intelligence® | Q1 2026, selected European markets

Romania, however, is an exception to this rural story, with Starlink sample share below 1% despite a large rural population and a national fixed median download speed of 283.06 Mbps. This reflects a history of heavy investment and dense fiber penetration in Romania led by fiber-first DIGI, which has delivered gigabit speeds to a very large share of the population at very low retail prices.

Spain, Europe’s fiber darling, also shows how strong fixed networks and targeted public satellite programs can limit Starlink’s mass-market role. Spain’s national median download speed was 277.98 Mbps in Q1 2026, 110.24 Mbps faster than Starlink. Its Conectate35 satellite program offers rural satellite broadband where fixed terrestrial coverage is missing, which makes satellite more of a targeted coverage tool than a mainstream substitute.

Europe’s LEO broadband market is broadening beyond Starlink

While Starlink is the most visible residential low Earth orbit broadband provider in Europe, it is part of a wider satellite connectivity market. Eutelsat OneWeb operates a 600-plus satellite LEO network focused more on enterprise, government, mobility, and backhaul. Amazon Leo began full-scale deployment in April 2025 and had more than 300 satellites deployed by late April 2026, although it has yet to appear as a European residential broadband force in our Speedtest data.

Europe is also building a sovereign satellite layer through IRIS², which is designed around secure, sovereign-controlled government connectivity. These services are different, but together they show why satellite broadband is moving beyond a niche conversation about one dish on one remote home.

Satellite is the latest tool in the broadband delivery suite

The consumer takeaway is straightforward. Starlink is becoming a credible home broadband option in parts of Europe where fixed networks are weak, delayed, expensive to extend, or needed as backup. It is strongest where the last few percent of premises are hardest to serve.

It remains a complement to fiber-led policy, with a smaller role in mature fixed markets. Where fiber, cable, or fixed wireless are fast, affordable, and widely adopted, Starlink remains a smaller part of the market. Where those networks thin out, the Starlink data shows where Europe’s hardest broadband problems still live.

Stay tuned for a follow-up analysis that will examine where LEO broadband fits in Europe’s long-term connectivity mix, including rural subsidy design, backup and resilience use cases, routing performance, and the role of Starlink and other satellite providers in serving the hardest to reach premises.

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.

| May 4, 2026

Europe's Hidden Mobile Performance Gap: Peak-Hour Congestion and Seasonality

Evening performance drop exposes the congestion problems telecom policy still misses.

The standard way of evaluating mobile network quality in Europe still leans heavily on aggregate metrics. National median speeds, coverage percentages, and 5G adoption rates are useful, but they flatten the hour-by-hour load profile that determines how networks feel when demand is highest.

Across the 30 markets in this analysis, the most consistent trough in download performance appears between 19:00 and 21:00 local time. We use that window as the evening peak and compare it with 02:00 to 05:00 local time, when demand is lowest. The difference between those windows captures a practical form of congestion: how much performance is lost when shared radio, backhaul, and core resources are under pressure.

This analysis draws on consumer-initiated Speedtest® samples across all 27 EU member states plus Norway, Switzerland, and the United Kingdom during Q1 2026, with trend and seasonality views extending from January 2024 through March 2026. For this article, we developed a peak-hour congestion framework that combines five dimensions of degradation: median download speed loss, loaded latency inflation, queue growth, jitter increase, and the decline in 10th percentile download speeds. The higher the value on the 0 to 100 scale, the more severe the measured peak-hour degradation.

Key Takeaways:

  • Spain is Europe’s most congested mobile market at evening peak, with a framework value of 62. Median download speed fell from 161.20 Mbps off-peak to 54.10 Mbps during peak hours in Q1 2026, a 66% drop, while loaded latency increased 60% to 724 ms.
  • Six markets maintained near-flat daily performance. Luxembourg (~0), Belgium (2), Norway (8), Slovakia (8), France (11), and the Netherlands (12) sit in the resilient tier, each with distinct structural characteristics across data-usage intensity, population mobility, and network density that help mitigate congestion.
  • Switzerland is the clearest example of why headline metrics alone are insufficient. Despite having Europe’s highest mobile ARPU at €50.90 (US$59.58) per subscriber and a 74% 5G connection share, Switzerland has the third-highest congestion value in the analysis at 47. Its median speed drop is moderate, but loaded latency rises 46% and the bottom 10% of users see download speeds fall 81%, from 25.50 Mbps to 4.80 Mbps.
  • Investment intensity and network management explain more than wealth, spectrum holdings, or market concentration. Capex as a share of revenue shows the strongest relationship with congestion resilience among the structural variables tested, although it is a moderate relationship rather than a deterministic rule. Operator gaps reinforce the point: in Poland, the evening-peak gap between T-Mobile and Plus is 4.1x, compared with 2.2x off-peak, meaning peak load can amplify rather than merely reflect baseline differences.
  • 5G improves the experience under load, but it does not remove congestion. Across 10 high-5G European markets, the average speed drop at peak is 32% for 4G and 27% for 5G. The more consistent 5G advantage is latency: 5G loaded latency at peak is 12% to 44% lower than 4G in every market tested.
  • Seasonality materially changes the congestion picture. Spain and Croatia show repeated summer pressure linked to tourism, Nordic markets show a summer shift toward rural and holiday-home locations, while Switzerland and Austria see congestion ease in summer, pointing to winter demand concentration at ski resorts as the sharper stress pattern.

Network Congestion Is a Regulatory Blind Spot

Mobile networks operate over a shared radio medium where spectrum is finite and the capacity of each cell sector is bounded by spectral efficiency, antenna configuration, interference management, and backhaul dimensioning. Unlike fixed broadband, where each subscriber typically has a dedicated last-mile connection, every mobile user in a cell sector draws from the same pool of radio resources.

When simultaneous demand exceeds what the available spectrum, radio configuration, and transport layer can deliver, per-user throughput falls, latency increases as queues build in network buffers, and the experience of every user on that sector deteriorates in tandem. This is why congestion is not just a speed issue. It is also a latency, consistency, and worst-user issue.

The anatomy of a single mobile cell at off-peak versus evening peak, showing shared spectrum, queue buildup, throughput compression, and tail-user collapse

The challenge is compounded by the geographic unpredictability of mobile demand. Operators must dimension networks for the busiest hour of the busiest day, even though average utilization is far lower. They must also do so across thousands of sites where traffic patterns shift with commuter flows, events, tourism, and seasons.

Despite this, most regulatory benchmarks and national performance reports still do not distinguish clearly between off-peak and peak-hour outcomes. The EU’s Digital Decade targets specify gigabit networks for all households and 5G coverage for all populated areas by 2030, but they do not set a comparable benchmark for performance under load.

BEREC’s 2024 implementation report on geographical surveys of network deployment also illustrates the difficulty. Expected peak-time speed is treated as one of the more challenging indicators for regulators to collect and standardize, and mobile quality-of-service reporting remains uneven across markets. The European Commission’s proposed Digital Networks Act may help simplify investment conditions, but it does not remove the need for better evidence on how networks perform during the hours of greatest demand.

Profiling Congestion Requires Looking Beyond Headline Speed

The congestion framework used for this article combines five dimensions of peak-hour degradation, each capturing a different facet of user experience. Throughput loss, weighted at 30%, measures the drop in median download speed from off-peak to peak. Loaded latency inflation, also weighted at 30%, captures how much delay increases during active data transfer, a direct indicator of network queuing that affects video calls, gaming, interactive web browsing, and increasingly AI-enabled real-time applications.

The five components of the peak-hour congestion framework — speed drop 30%, latency under load 30%, buffer pressure 20%, stability decay 10%, worst-served users 10%

Queue growth, weighted at 20%, isolates congestion from baseline network quality by measuring how the gap between idle and loaded latency widens. Jitter inflation, weighted at 10%, reflects the stability degradation that impairs real-time communication. The 10th percentile download drop, weighted at 10%, captures how much the worst-served users suffer, which is especially relevant to policy debates about universal service quality.

Loaded latency is particularly important. A network can maintain superficially reasonable throughput while loaded latency rises from 400 ms to 700 ms or more, degrading video calls, increasing application response lag, and creating a perceptibly worse user experience that median speed alone does not reveal.

A Wide Peak-Hour Gap Separates Europe’s Best and Worst Mobile Markets

The 30 markets analyzed segment into four tiers when applying the congestion framework used for this research. The top and bottom of the distribution are not separated by marginal differences. Spain’s framework value of 62 is more than five times the Netherlands’ 12 and roughly eight times Norway’s 8.

Spain Tops Europe's Peak-Hour Congestion Severity by a Wide Margin
Speedtest Intelligence® | Q1 2026

Six markets are congestion-resilient: Luxembourg, Belgium, Norway, Slovakia, France, and the Netherlands. These markets maintain near-flat performance profiles across the day. The Netherlands delivers 157.90 Mbps at evening peak, just 15% below its off-peak level. Norway’s loaded latency varies by fewer than 70 ms across the 24-hour cycle.

Belgium and Luxembourg show speed gains, meaning evening peak speeds actually exceed their nighttime baseline, likely reflecting business-hour demand relaxation (unsurprising in Luxembourg where many commute into and out of the country each day for work) and, in some cases, overnight energy-saving configurations that reduce available radio capacity (i.e., disabling higher bands and features like higher order carrier aggregation) during the off-peak reference window.

Europe's 24-hour mobile heartbeat across selected European markets, showing the synchronized evening trough

Eleven markets fall into the moderate tier. Speed drops here range from around 30% to more than 45%, but absolute peak performance varies significantly, from Bulgaria’s 142.80 Mbps to Romania’s 62.10 Mbps. Germany, Europe’s largest mobile market by revenue, sits in this tier with a 34% speed drop and a congestion trajectory that has been quietly worsening.

Ten markets show significant congestion. Italy, hosting the EU’s most fragmented mobile market structure (by HHI concentration), delivers just 45.20 Mbps at peak, the lowest absolute peak speed of any major EU economy in the analysis. The Herfindahl-Hirschman Index (HHI) is a measure of market concentration: lower values indicate a more fragmented (or competitive) market structure. This potentially reflects the real-world network quality costs imposed by the market’s historical focus on price competitiveness.

Three markets face severe congestion: Switzerland, Ireland, and Spain. All three are three-operator markets (although DIGI is building a fourth network in Spain) and all three feature below-average capex intensity. Ireland and Spain also combine low to medium ARPU, high mobile data usage, and widespread unlimited or near-unlimited tariffs, which likely contribute to higher load pressure per subscriber despite high FTTH penetration.

Peak-hour congestion framework values across 30 European markets, Q1 2026 — Spain, Ireland, and Switzerland in the severe tier

The three Benelux markets form a notable cluster at the resilient end of the scale. Their shared characteristics, including small and dense geography, high urbanization, strong fixed broadband penetration supporting Wi-Fi offload, mature three-operator market structures (changing as DIGI becomes a fourth operator in Belgium), and less exposure to national-scale seasonal coastal tourism, appear to create structural conditions that resist congestion.

Speed Rankings Alone Disguise Severe Latency Degradation in Europe’s Wealthiest Markets

Switzerland’s congestion outcomes challenge several assumptions about what makes a well-performing mobile market. It features the highest mobile ARPU in Europe at €50.90 (US$59.58) per subscriber (based on GSMA Intelligence data), the highest 5G connection share at 74%, and 99% reported outdoor 5G population coverage. In aggregate speed terms, Switzerland would not look like an obvious congestion outlier.

Under the congestion framework, however, Switzerland ranks third-worst in Europe with a value of 47. The headline speed drop of 36% appears moderate. But loaded latency inflates 46% at peak, and the bottom 10% of Swiss users experience an 81% collapse in download speed, from 25.50 Mbps off-peak to 4.80 Mbps at peak. This 10th percentile collapse is the worst of any market in the analysis, meaning the most vulnerable Swiss mobile users, likely those in congested urban cells or at the edge of coverage, effectively lose functional mobile broadband during evening hours.

Each European market's evening-peak failure mode — speed loss versus latency inflation, with severely congested markets clustering in the upper-left quadrant

Operator-level data identifies the specific source of the problem. Sunrise, which holds approximately 27% of the Swiss mobile market with 3.1 million mobile customers, shows a 73% speed drop at peak, falling from 164.00 Mbps off-peak to 44.50 Mbps. Its loaded latency inflates 57% and its 10th percentile download speed falls to 3.10 Mbps. Swisscom, operating in the same geography with approximately 54% market share, drops 31% and maintains 97.90 Mbps at peak with a 10th percentile download speed of 10.60 Mbps. Salt, the third operator, falls between the two with a 41% speed drop.

The difference is not simply that Swisscom is faster in general. Off-peak, the gap between the fastest and slowest Swiss operator is only 23.40 Mbps, or 1.17x. At peak, the gap expands to 53.40 Mbps, or 2.2x. Evening demand therefore exposes an operator-level resilience gap that is mostly hidden overnight.

Spectrum holdings provide part of the explanation. Swisscom holds 743 MHz of total assigned spectrum, including 613 MHz of mid-band capacity across the 1500, 1800, 2100, and 2600 MHz bands. That is roughly 2.7x the mid-band depth available to Sunrise (224 MHz) or Salt (220 MHz). Because Swisscom also serves a larger customer base, that advantage is less dramatic on a per-subscriber basis, but it remains directionally favorable. The fact that Salt has broadly comparable mid-band depth to Sunrise yet manages a materially better peak outcome suggests that deployment, traffic mix, site configuration, and network management matter alongside raw MHz.

Switzerland also presents a useful caution on investment interpretation. Its capex-to-revenue ratio is the lowest in the analysis at approximately 10% (based on GSMA Intelligence data), but absolute capex may look less weak because Swiss ARPU is high. The ratio still matters because it measures reinvestment intensity: how much of a high-revenue market is being put back into capacity.

Loaded Latency Reveals a Different Map of European Mobile Stress
Speedtest Intelligence® | Q1 2026

Regulation may also contribute. Switzerland’s non-ionizing radiation rules are more precautionary than the international exposure limits used in many other markets, and new or modified antenna installations must demonstrate compliance. These rules do not explain the Sunrise-Swisscom gap on their own, but they can raise the practical complexity of densification and capacity upgrades. The combination of high ARPU, low reinvestment intensity, strict site constraints (forcing high grid density), and large operator-level dispersion points to a market where headline metrics mask material quality-of-experience gaps that only become visible under demand pressure.

Intra-Market Differences Can Exceed Inter-Market Gaps

Our operator-level analysis shows that congestion outcomes within a single country can diverge more sharply than outcomes between countries. Four markets illustrate different patterns.

Spain, for example, shows a high-ceiling, high-collapse pattern. Orange, operating as part of MasOrange following the 2024 merger with MasMovil, delivers 329.40 Mbps off-peak, among the fastest off-peak speeds recorded for any operator in any market in this analysis. By evening peak, this falls 72% to 91.20 Mbps, with the 10th percentile dropping 91%. The raw network capacity demonstrably exists. The challenge appears to be distributing that capacity under concentrated evening demand, a pattern consistent with the complexity of post-merger network integration and traffic migration.

Movistar starts from a more moderate off-peak level of 120.00 Mbps but drops just 26% and maintains 89.20 Mbps at peak. Vodafone Spain shows the weakest absolute peak performance at 27.30 Mbps, with loaded latency reaching 1,189 ms.

Spain's Operator Performance Diverges Sharply Under Peak Load
Speedtest Intelligence® | Q1 2026

Poland shows an investment-divergence pattern. T-Mobile delivers 99.50 Mbps at peak with a 10th percentile download speed of 11.80 Mbps. Plus manages 24.30 Mbps with a 10th percentile of 1.90 Mbps. The 75.20 Mbps gap between operators serving the same country is the largest intra-market spread in our analysis. Crucially, the off-peak gap is much smaller proportionally: T-Mobile is 2.2x faster than Plus off-peak, but 4.1x faster at peak. That means the result is not merely a static speed hierarchy (i.e., peak demand amplifies the gap).

Poland’s congestion outcomes are also improving overall, with evening peak speeds up 35% year-on-year, largely driven by the T-Mobile and Orange networks and by the recent launch of mid-band 5G.

Peak demand doesn't always widen the operator gap — sometimes it shrinks it. Off-peak versus evening-peak operator speed ratios across seven European markets

Ireland, by contrast, shows a shared-ceiling pattern. Three, Vodafone, and Eir diverge widely off-peak, ranging from 99.20 Mbps to 167.00 Mbps. At peak, all three converge within a 13.80 Mbps band, between 34.60 Mbps and 48.40 Mbps. This convergence pattern is unusual among the operator markets analyzed and points to a structural capacity ceiling rather than one operator underperforming in isolation. Ireland’s three-operator market, high per-connection data usage, and low collective capex-to-revenue ratio (atop a rural-skewed geography) appear to create conditions where no operator can easily break away from the market-wide evening constraint.

Portugal, meanwhile, exhibits a deterioration pattern. The country’s evening-to-night performance gap widened from 11% to 34% between Q1 2025 and Q1 2026, the fastest deterioration in our analysis. The primary driver at the operator level is MEO, where peak 10th percentile download speed has fallen to 1.40 Mbps, the lowest figure recorded for any major operator in our European operator sample. This effectively represents a loss of functional service for MEO’s worst-served users during peak hours.

DIGI, which launched as Portugal’s fourth MNO in November 2024, shows a 25% speed drop with near-zero latency inflation of 7%. That result is consistent with the low utilization expected from a new entrant still building its customer base, rather than evidence of superior network engineering at full market scale.

5G Raises the Speed Ceiling but Does Not Prevent It From Being Hit

A persistent assumption in regulatory and industry discourse is that 5G deployment will resolve capacity constraints. Our data offers a more nuanced picture.

Across 10 European markets with significant 5G adoption, we segmented Speedtest® results by device-reported connection type. The average speed drop at peak is 32% for 4G and 27% for 5G. In absolute terms, 5G is substantially faster. A 5G user in Spain still receives 106.40 Mbps at peak versus 20.30 Mbps for a 4G user in the same market.

The proportional pattern, however, varies by market. In France and Norway, 5G peak speeds are actually higher than the 5G off-peak baseline. In Denmark and Switzerland, the proportional 5G speed drop is steeper than the 4G drop. The broad conclusion is therefore not that 5G removes congestion but that it raises the performance ceiling and often softens the evening decline, but it remains exposed to shared capacity constraints.

Peak-hour 4G versus 5G comparison across 10 European markets — 5G's most consistent advantage at evening peak is loaded latency, not the proportional speed drop

The more consistent 5G advantage lies in latency under load. In every market tested, 5G loaded latency at peak is lower than 4G, by margins ranging from 12% in Denmark to 44% in the United Kingdom. The U.K. contrast is the starkest. 4G users experience 904 ms loaded latency at peak, while 5G users experience 507 ms. This gap means congested 5G still materially outperforms congested 4G for applications sensitive to delay, including video conferencing, cloud gaming, interactive browsing, and emerging live voice and video AI applications.

This distinction matters for how policymakers and operators frame the 5G value proposition. 5G deployment expands the performance ceiling and delivers a real latency improvement that persists under congestion. But it should not be conflated with congestion resilience. A market can achieve high 5G adoption and still rank among Europe’s most congested. The variables that determine whether peak-hour performance holds, as mentioned earlier, are a combination of capacity investment, densification, spectrum deployment depth, backhaul dimensioning, and traffic management, not the generation label attached to the radio interface.

Seasonal Travel Shifts Europe’s Mobile Congestion Patterns

Analysis of monthly Speedtest® data from January 2024 through March 2026 shows that congestion is not static. It follows seasonal rhythms that differ sharply by geography. This long window allows two summers, two winters, and Q1 2026 to be compared.

Our seasonality analysis uses broad evening and nighttime windows rather than a single hour, reducing sensitivity to daylight-saving changes and one-off hourly effects. The metric here is the ratio of evening download speed to nighttime download speed. Lower values indicate a larger evening gap.

Three seasonal patterns emerge. In several markets, congestion worsens materially in summer. Spain shows the most extreme swing. The evening-to-night speed ratio fell from 60% in January 2024 to the low teens during summer 2024, then remained much weaker in July and August 2025 than in winter.

This aligns with Spain’s position as one of Europe’s most-visited countries. Spain welcomed 96.8 million international tourists in 2025, with a large share of arrivals concentrated in the summer months. These visitors are disproportionately mobile-dependent because they lack residential Wi-Fi offload, and they cluster in geographically constrained coastal zones.

Europe's seasonal congestion fingerprints — monthly evening-to-night download speed ratios from January 2024 through March 2026, grouped by pattern

Croatia shows an even more precise seasonal signature. Evening peak speed fell from 58.70 Mbps in January 2024 to 34.90 Mbps in August 2024. The pattern repeated in 2025, with evening speed falling from 71.60 Mbps in June to 35.30 Mbps in August. Croatia recorded 4.7 million tourist arrivals and 27.2 million tourist nights in commercial accommodation in August 2024, a major seasonal load for a country with a resident population of roughly 3.9 million. The concentration of tourism along the Adriatic coast creates acute demand pressure on a relatively narrow cellular footprint.

Nordic markets show a different summer pattern driven less by inbound tourism than by domestic movement toward second homes and rural leisure areas. Norway’s evening peak speed dipped to 77.10 Mbps in July 2024 and 102.40 Mbps in July 2025, compared with 121.40 Mbps and 130.70 Mbps in the respective January periods. Norway has a large stock of holiday homes, many in low-density areas where cellular capacity is designed around lower year-round demand. When urban populations move to these areas during summer, demand shifts toward cell sites that may not be dimensioned for short seasonal peaks. Denmark, Sweden, and Finland display related patterns tied to summer-house traditions.

A final group moves in the opposite direction. In Switzerland, the evening-to-night speed ratio improved from 44% in January 2024 to 76% in August 2024, and from 63% in January 2025 to 85% in August 2025. Austria shows a similar, though less pronounced, pattern.

This points to winter demand concentration as the sharper stress period, likely reflecting a combination of indoor usage, tourism in ski regions, and more difficult terrain for capacity planning.

Investment Intensity Is the Better Indicator of Congestion Resilience

To test which structural factors may shape congestion outcomes, we compared the framework values against market variables drawn from GSMA Intelligence, national statistical authorities, and public data sources.

Our results challenge several common assumptions. National wealth does not explain congestion well. GDP per capita has only a weak negative relationship with measured congestion. For example, Austria, with a GDP per capita of €49,777 (US$58,269; per World Bank data), carries a congestion value of 37, while Romania, at €17,154 (US$20,080), records a lower framework value of 28.

Mobile ARPU tells a similarly mixed story. Higher ARPU appears to support higher absolute peak speeds, but it does not determine whether those speeds hold under peak demand. Switzerland has Europe’s highest mobile ARPU and still ranks third-worst under our congestion framework. ARPU can fund capacity, but it only improves resilience when revenue is actually converted into spectrum deployment, site upgrades, densification, and transport capacity.

Spectrum holdings also require care. Total spectrum per operator shows only a weak relationship with congestion outcomes, and mid-band spectrum per operator shows almost no relationship in this dataset. Spectrum enables capacity, but it does not create capacity on its own. It must be deployed, sectorized, integrated with backhaul, and matched to traffic demand. This is where cell site density likely matters.

The strongest structural relationship we found is capex as a share of revenue. In plain terms, markets where operators reinvest a larger share of revenue tend to hold up better at peak, although the relationship is moderate rather than absolute. Norway, at 24% capex-to-revenue, records a framework value of 8. Switzerland, at 10%, records 47. Both are small, wealthy, three-operator markets with high ARPU. The difference is not simply that one has more money available. It is that one reinvests a larger share of revenue into the network (but also, importantly, has a less intense usage profile).

Market concentration, measured by the Herfindahl-Hirschman Index, shows a weak and counterintuitive negative relationship with congestion. More concentrated markets are not necessarily worse. Italy, the most fragmented mobile market in our sample by this measure, carries a framework value of 41 and the lowest absolute peak speed of any major EU economy at 45.20 Mbps. The Netherlands, among the more concentrated markets with three operators, records 12 and delivers 157.90 Mbps at peak.

Rural population share shows a moderate positive relationship with congestion and the strongest relationship in our dataset with 10th percentile performance. More rural countries systematically deliver weaker outcomes for the most poorly served users at peak (likely contributing to Ireland’s weak standing, for instance), consistent with the challenge of dimensioning capacity across dispersed populations and more extensive coverage footprints.

Peak-Hour Performance Should Become a Regulatory and Competitive Benchmark

The gap between what European mobile networks can deliver under light load and what they provide during the hours of highest demand is material, measurable, and largely invisible to most public benchmarks.

The trajectory of Speedtest® data offers cautious grounds for optimism in some markets. Ireland’s evening peak speed improved from 20.90 Mbps in Q1 2025 to 47.00 Mbps in Q1 2026, a 125% gain (reflecting diversified spectrum deployment post-auction). Poland improved 35% over the same period, reflecting the early impact of mid-band 5G rollout. The U.K. improved 18%, a trend consistent with early network-integration effects following the Vodafone-Three merger, which completed on 31 May 2025.

Year-on-Year Trajectory Splits Europe Into Improvers and Decliners
Speedtest Intelligence® | Q1 2025 vs Q1 2026

But these gains coexist with deterioration elsewhere. Portugal’s evening-to-night performance gap widened from 11% to 34% over 12 months, a 23 percentage point increase. Germany’s widened from 20% to 29%, a 9 percentage point increase, even though its evening speed improved slightly. In Germany’s case, nighttime performance improved faster than evening performance, widening the gap that consumers experience between low-load and high-load hours.

Congestion is not an inevitable consequence of demand growth (which itself is slowing in mature markets). Countries with sustained mobile investment intensity, well-managed spectrum deployment, sufficient densification, and enough revenue to fund capacity demonstrate that peak-hour performance can be maintained even as traffic grows or spikes shift.


Methodology

This analysis draws on Speedtest® data from consumer-initiated mobile Speedtest measurements. The primary snapshot covers Q1 2026, January through March, across all 27 EU member states plus Norway, Switzerland, and the United Kingdom. Trend and seasonality analysis extends from January 2024 through March 2026.

Peak hours are defined as 19:00 to 21:00 local time, confirmed as the consistent trough across markets by examining full 24-hour performance profiles. The off-peak baseline is defined as 02:00 to 05:00 local time. The off-peak period is not intended to represent normal consumer usage. It is a low-load reference window used to estimate what the network can deliver when demand pressure is minimal. However, the off-peak baseline should be interpreted as a low-load observed baseline, not necessarily a maximum engineering-capacity baseline, because some networks may apply overnight energy-saving configurations that reduce available radio capacity.

The peak-hour congestion framework combines five components: 30% median download speed drop, 30% loaded latency inflation, 20% queue growth, 10% jitter inflation, and 10% 10th percentile download speed drop. Higher values indicate more severe measured peak-hour degradation.

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.

| April 30, 2026

The Iberian Blackout, One Year On: Operators Are Outpacing Regulators on Telecom Resilience

The Iberian blackout showed how quickly mobile networks can follow the power grid down.

On 28 April 2025, at 12:33 CEST, the largest stress test of telecom infrastructure in modern European memory unfolded in less than three hours. A cascading collapse of the Iberian power grid severed mains electricity to virtually all of mainland Spain and Portugal, forcing mobile sites onto battery and generator backups and triggering a near-synchronous collapse of mobile coverage across the peninsula. By the late evening, more than half of mobile users in large parts of Spain and over 60% in the worst-affected areas of Portugal had no service at all.

Twelve months on, the Iberian regulatory and operator response has not converged. Spain has drafted Europe’s most ambitious telecom autonomy rule but has not enacted it. Portugal’s parliament published cross-party recommendations on the anniversary week. And operators, in the meantime, have made the most concrete commitments of all.

Key Takeaways:

  • Spain has drafted, but not enacted, the EU’s most ambitious telecom autonomy rule. The 24/12/4-hour tiered draft Royal Decree, with an 85% population coverage target including 112 calls, remained at consultation stage on the anniversary, twelve months after the blackout.
  • Portugal’s response is broader. ANACOM has recommended minimum autonomy for network elements, Cell Broadcast, satellite redundancy, and real-time network visibility, while the government has paired telecom recommendations with electricity and critical-infrastructure measures.
  • Operator investment is now part of the policy equation, from Vodafone’s Enhanced Power program to MEO/Altice route-diversity projects. Our data showed that these are the right investment targets; operators with wider or deeper backup-power layers flattened and delayed the outage curve (e.g., MEO), while thinner autonomy produced fast, severe loss of service (e.g., DIGI).
  • Satellite has emerged as a working fallback layer in Iberia, not a theory. Starlink rerouted Spanish gateways through Madrid, London and Milan during the blackout itself, and Portuguese user activity ran roughly 196% above baseline during Storm Kristin in February 2026. This strengthens the case for LEO satellite in emergency kits, government continuity plans, and remote-site backup, but it is not a panacea on its own.

Spain has drafted, but not enacted, the EU’s most ambitious telecom autonomy rule

Spain has gone furthest toward converting the blackout into explicit telecom rules. In December 2025, the Ministry for Digital Transformation opened consultation on a draft Royal Decree on the security and resilience of electronic communications networks and digital infrastructure. The draft classifies telecom networks and selected digital infrastructure as essential in emergencies and applies to telecom operators as well as certain submarine cable, satellite, data center, and internet exchange assets.

The most important shift is that resilience is specified in hours. First-level infrastructure would need at least 24 hours of operation during a power interruption, intermediate sites at least 12 hours, and other sites four hours. For mobile networks, the four-hour requirement must maintain coverage for 85% of the population, with operators able to prioritize voice over data or critical services over higher-capacity layers. The decree also strengthens 112, public-alert, incident-reporting, and coordination obligations.

This is notably sharper than a generic critical-infrastructure designation. It forces operators to classify sites, power paths, and services before the crisis. It also surfaces implementation questions around rooftop weight limits, generator fuel logistics, RAN sharing, and rural refueling. The CNMC’s formal opinion on 12 March 2026 recommended progressive rollout, prioritizing “cost-efficient solutions like inter-network roaming and satellite backup,” and full alignment with NIS2.

Both observations cut to the heart of a regulatory landscape Spain itself has not finalized: the country has not transposed NIS2 (EU cyber-resilience law for essential sectors) and was one of 19 member states to receive a Commission reasoned opinion on 7 May 2025. The cost gap between the government’s €73 million (US$85 million) estimate and operator estimates closer to €300 million (US$351 million) for power hardening has not narrowed. Public consultation closed on 8 January 2026 but, as of the anniversary, the decree had not been promulgated. Energy Minister Sara Aagesen, appearing before the Senate on 23 March 2026, was still asking power companies to publish the underlying outage data.

Portugal’s parliament has set a more operational and structural agenda

Portugal’s response has been less prescriptive at the telecom-site level, but broader across emergency systems. A government summary of the ANACOM blackout report said the prolonged power failure cascaded across fixed and mobile networks, affected 112 access, and constrained emergency communications.

The regulator’s own published recommendations reframed resilience as an operational checklist: minimum autonomy times for batteries and generators, renewable extensions at cell sites, restricted SIM/eSIM access for authorities, 112 routing diversification, and an evolution of public warnings towards Cell Broadcast (the standardized 4G/5G capability to push warning messages to all phones in a defined cell area without an app or sign-up).

The Cell Broadcast recommendation is important. Unlike SMS, Cell Broadcast sends one-to-many alert messages to compatible phones in a targeted radio area, making it better suited to fast public warnings and less exposed to congestion. It still needs powered cell sites, but it is a better emergency-warning tool than individual SMS once networks degrade.

The most consequential update in Portugal, however, came in the anniversary week itself. The cross-party parliamentary working group led by PSD deputy Paulo Moniz published its final report on 23 April 2026, recommending 72 hours of energy autonomy at hospitals, primary health centers, nursing homes and emergency services; 24 hours at all other critical infrastructure; the formal classification of food retailers and pharmacies as critical; and, most pointedly for telecoms, a national emergency alert system independent of commercial mobile networks. The structural separation of public-safety communications from operator-controlled infrastructure is the most ambitious year-two recommendation in either country.

Portugal also had a real-world second test. Between late January and February 2026, Storms Kristin, Leonardo and Marta cut median mobile download speeds by 52% at peak in Speedtest data, with around 40% of telecom failures linked back to electricity loss at mobile sites. ANACOM activated national roaming and Starlink user activity surged ~196% above baseline based on our analysis (with authorities also deploying LEO terminals to remote areas). The episode did not so much reopen Portugal’s policy debate as confirm it.

Operators have outrun both governments

The most concrete Iberian telecom resilience commitments of the past year have come from operators rather than regulators. Vodafone Group’s Enhanced Power initiative, announced on 28 November 2025 with Portugal as its first deployment region, will reinforce more than 10,000 critical sites Europe-wide. Tier specifications include 72 hours of backup or guaranteed refueling within 48 hours at more than 400 mobile data centers, and four hours at aggregation and critical access sites. The initiative is supported by an AI-controlled adaptive power-backup function already live in Greece and in trial in Turkey, which the company says nearly doubles base-station battery duration in certain scenarios.

The day of the blackout itself drew the operator distinction more sharply than any policy document could. MEO’s network peaked at just over 16% of subscribers without service, the best performance observed across either country, because deeper, more widely deployed power reserves materially flattened and delayed its outage curve. NOS peaked at around 30%; Vodafone Portugal at nearly 70%. DIGI’s still-nascent network exhibited a near step-function collapse, with up to 90% of subscribers without coverage for more than a day, pointing to gaps in core network geo-redundancy as well as site-level backup.

Building on this, more recently, Portuguese reporting highlighted MEO’s expansion of its Linda-a-Velha network center and a new Porto landing station intended to improve transatlantic route redundancy. Cellnex, Spain’s tower incumbent and a named addressee of the country’s draft decree, meanwhile reported an integrated approach to UPS, generators and battery banks across more than 120,000 sites in 10 countries, although per-site backup-hour disclosures remain opaque.

Next steps: convert resilience scaffolding into rules

The Iberian blackout exposed three layers of telecom exposure simultaneously: power autonomy at the access network, cross-border interconnection at both energy and transport layers, and the dependence of upstream services on landing stations and gateways that are themselves geographically concentrated. Year one has produced movement on all three, but unevenly.

The European Commission’s proposed Digital Networks Act may help by simplifying rules, supporting satellite services, and improving security cooperation. It will not, by itself, set a uniform backup-power floor for mobile sites. Iberia’s lesson is that national implementation still matters.

Spain has the draft, but a draft is not a rule. Promulgating the December 2025 Royal Decree, or phasing it in along the progressive lines the CNMC has suggested, is the single highest-leverage regulatory move available to either Iberian government in the next twelve months. Portugal’s parliamentary recommendations need to translate from cross-party report to executive action; the proposed national emergency alert channel that does not depend on commercial mobile networks is the most structurally significant call in either country, and would change how public-safety messaging routes through Iberian telecom infrastructure in a future event.

Operators will keep moving voluntarily. But the next phase of resilience policy in Iberia will only be tractable if disclosure quality improves to make per-site backup-hour benchmarking comparable across networks. The same lesson is being drawn in markets that have faced very different stress tests: our analysis of Cyclone Alfred’s impact on Queensland networks reaches the same conclusion from the opposite hemisphere. What gets measured, gets hardened.

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.

| April 29, 2026

Starlink-powered O2 Satellite Puts the UK at the Front of Europe’s D2D Race

Early Ookla signal scan data shows broad but shallow uptake across UK not-spots, as Ofcom’s first-mover framework turns the market into Europe’s direct-to-device testbed

The UK has become the first country in Western Europe where a smartphone can reach a satellite directly without specialist hardware or a separate app. Virgin Media O2 launched O2 Satellite on February 26, 2026, using SpaceX’s Starlink Mobile constellation over licensed 1800 MHz spectrum, and priced it at £3 per month on Pay Monthly plans or at no extra cost on high-end “Ultimate” tariffs. The launch follows a partnership announced in October 2025 and an Ofcom authorization framework finalized in December 2025.

Analysis of Ookla’s background signal scan data captures that launch from the real-world handset side. Between July 2025 and March 2026, the number of unique monthly UK users registering with a direct-to-device (D2D) satellite rose from negligible early levels to a clearly visible footprint, an order-of-magnitude shift that has already lifted the U.K. from a rounding error in our global D2D dataset to the world’s third-largest market by unique-user count by March 2026, behind only the United States and Australia and ahead of Canada, Chile, and Peru.

Key Takeaways:

  • The UK has quickly become one of the largest detected D2D markets globally, but not one of the deepest. By March 2026, the UK ranked third among countries with live commercial D2D services by detected D2D users, behind the U.S. and Australia, but only eighth by D2D user share and ninth by D2D scans per detected user. In the U.K., the dominant pattern appears to be many users briefly crossing into satellite-eligible conditions, with relatively few remaining on D2D for extended periods.
  • D2D utilization within the UK mobile base reached 0.30% in March 2026 (based on the share of all mobile users coming from D2D), and scans per D2D user averaged four, well below Canada’s 29. The utilization rate matches the US nine months into its T-Mobile service, and within VMO2’s addressable base utilization rises to approximately 1.4%, in line with more mature D2D markets.
  • Geographic concentration tracks the UK’s known coverage gaps. From November 2025 onward, we observed D2D samples lighting up across the Scottish Highlands, the Outer Hebrides, the Welsh uplands, the Southwest peninsula, and the North York Moors, the same areas Ofcom’s Connected Nations 2025 report identifies as all-operator coverage blackspots and that the £1.3 billion Shared Rural Network is committed to closing by January 2027.
  • A multi-operator D2D split is already forming in the UK. VodafoneThree received Ofcom’s second D2D license variation on April 15, 2026 on 900 MHz (Band 8), paired with AST SpaceMobile’s broadband-capable BlueBird satellites through Satellite Connect Europe. BT/EE has not yet announced a D2D handset service and is instead prioritising fixed Starlink broadband. It also appears best placed to support any satellite layer for the UK’s Emergency Services Network, the national communications platform being built to connect police, fire, ambulance and other first responders.

Methodological note: This analysis draws on nine months of Ookla signal scan data from UK Android handsets that registered at least once with a known D2D carrier network between July 2025 and March 2026. Coverage is limited to recent Samsung flagships, reflecting O2’s narrower device support versus more mature D2D services elsewhere. Our data shows O2 Satellite emerging in the data before its public launch on February 26, 2026, then scaling across the first full month of commercial availability.


O2 Satellite was already ramping before the public launch

Virgin Media O2 announced its Starlink Direct to Cell (D2C) partnership on October 30, 2025, describing O2 Satellite as a service that would initially provide messaging and data, work automatically in areas without traditional O2 coverage, and aim to raise O2 landmass coverage from 89% to more than 95% within 12 months of launch. The same announcement said internal trials were already underway.

Analysis of our background signal scan data offers the first clear empirical sign of a structured internal rollout becoming visible in passive measurement before any public announcement. UK detected D2D users were essentially negligible and flat between July and October 2025. In November, during the period in which VMO2 publicly noted that its employees were “already using the technology in real-world conditions across the country,” we captured a sharp inflection in usage, with activity moving well above the Q3 2025 baseline by the end of 2025.

UK D2D activity jumped twice before and after launch
Ookla Background Signal Scans | United Kingdom, July 2025 to March 2026 | Q3 2025 average = 1x

The second inflection landed in February 2026, when Ofcom granted VMO2 its first D2D license variation on 17 February. The Exemption Regulations came into force on February 25th, and the consumer service switched on the following day. As a result, detected D2D activity stepped up again in both February and March post-launch and is now scaling further.

The UK’s early D2D usage is broad, but still shallow

The UK ranked third globally by unique D2D users by March 2026 in our dataset, with 11% of the world’s tracked D2D user count, behind the U.S. at 37% and Australia at 14%. It sits ahead of Peru, Canada, Chile, Ukraine, New Zealand, Puerto Rico, and Japan. On scan volume the UK is sixth, a more modest figure that reflects how recently the O2 service launched and how little behavioral load each UK user is yet carrying.

UK ranks third by detected D2D user share, but scans-per-user remain low
Ookla Background Signal Scans | March 2026, Speedtest-derived

To clarify, a unique detected user means a device appeared on a D2D network (like Starlink Mobile) at least once in the month. Scans per detected user indicate how much repeat background activity (or usage of the D2D service) those detected users generated.

In the context of the UK data, this means the initial D2D service is appearing across a meaningful detected user base, but each detected user is generating relatively few satellite scans. For example, one user might take a long trip into the backcountry, generating lots of scans over the course of days. Another user might briefly link to a satellite during a walk in the country, generating only one scan. Both are considered unique D2D users, but one generates far more scans. That points to short or intermittent satellite registrations, a fallback service that appears at the edge of terrestrial coverage (which may be less likely to be encountered frequently in the U.K. compared to other large landmass countries due to the very high level of urbanization, see below), and a product that is still constrained by few supported devices (only the two most recent generations of Samsung flagships are supported), supported apps, outdoor use, and the requirement for an open view of the sky.

It would be premature to read scan depth alone as a direct measure of user engagement or paying customers at this early stage. Background scan cadence, device mix, app eligibility, operating-system behavior, and movement patterns all affect the data. Even so, the cross-country contrast is large enough to be analytically useful here. In the U.K., the dominant pattern appears to be many users briefly crossing into satellite-eligible conditions, with relatively few remaining on D2D for extended periods. Penetration within the mobile base (based on the share of all signal scans) reached 0.30% in March 2026, compared with 0.46% in the U.S., 0.70% in Canada, 1.26% in Chile, and 1.91% in New Zealand. That places U.K. penetration at roughly the level the U.S. had reached within its first nine months of service. Within VMO2’s addressable base, and assuming our background scans distribute across U.K. operators in approximate proportion to their mobile market share, penetration rises to approximately 1.4% in March 2026, closer to more established markets despite only six weeks of commercial service.

The contrast with the U.S. and Canada is instructive. Both those markets have seen D2D user counts fall since summer 2025, by 17% and 48% respectively through March 2026, coinciding with T-Mobile and Rogers ending their initial free-trial D2D tiers and moving to pricing gated by high-tier plans. VMO2 launched with a £3 bolt-on and free inclusion on its Ultimate tariffs from day one. The U.K. model therefore likely avoids the drop that free-trial expiration has produced in mature markets, though whether that pricing structure holds as the service scales beyond the early-adopter cohort is an open question. Seasonality is also likely to matter: outdoor travel and recreation typically dip in winter and should be watched again through spring and summer 2026.

The geospatial pattern shows D2D forming at the edge of O2’s network

Analysis of the geospatial distribution of the U.K. D2D usage shows three phases between November 2025 and March 2026. In November and December, the footprint was sparse and more concentrated in southern and eastern England (likely reflecting early testing), including areas closer to higher population density and travel corridors. By January and February, detections had spread more clearly into Wales, the Midlands, northern England, and Scotland (concentrated in the areas that Ofcom’s Connected Nations reports identify as the U.K.’s most coverage-constrained, which also happen to be popular for outdoor activities).

This pattern is important because it complicates the idea that D2D demand is only about the most remote places. O2 Satellite works where the main O2 network is unavailable and satellite coverage is available (notably excluding the major indoor cellular coverage gaps that persist), which means it can matter in partial not-spots as well as total not-spots. A partial not-spot can still be a real coverage gap for an O2 customer, even if another mobile operator has terrestrial service there. In practical terms, a hillwalking route, coastal road, or visitor-heavy rural area can be a D2D use case even when it is not a vast wilderness.

In Scotland, D2D usage clusters have been observed in Argyll and Bute, the Northwest Highlands, the Inner Hebrides, and the Outer Hebrides. Ofcom’s latest figures put Scotland at 89% 4G coverage from at least one operator and 65% from all four, the lowest in the U.K., and parts of the Highlands and Islands still sit materially below that average. These are the places where a satellite-to-phone overlay has the most work to do, and they show up distinctly in our data.

A second cluster runs through Wales, concentrating in Powys, Gwynedd, and mid-Ceredigion. A third picks up across the Southwest peninsula, particularly North Devon, Exmoor, Bodmin Moor, and West Cornwall. A fourth sits across the North of England, across the Yorkshire Dales, the North York Moors, Northumberland, and parts of the Lake District. The East Anglian coast and the Lincolnshire Wolds complete the rural pattern.

This is why the overlap with Shared Rural Network target areas matters. The SRN is the £1.3 billion joint program between the four U.K. operators and government designed to raise all-operator 4G coverage to 89.2% of U.K. landmass by January 2027, with Scotland and Wales carrying the largest share of the uplift. As of the 2025 update, Extended Area Service site deployments are contributing an additional 0.25% to 1% of UK landmass coverage across operators, and 40 Scottish SRN sites are now live.

Our data suggests that for many of the same locations, D2D is now delivering a first-layer (albeit highly constrained in its current form) connectivity experience well before the SRN timeline concludes. This helps explain why early U.K. detections can appear around populated and visitor-heavy regions rather than only in the least populated parts of the country. The relevant question is not simply where the U.K. is empty but where compatible O2 customers, outdoor use, travel patterns, coastal and upland activity, and O2-specific terrestrial gaps overlap.

O2’s own launch framing points to that same use case. The operator described the service as helping users when hiking, climbing, doing water sports, sailing, or traveling in rural, coastal, and remote locations. The early geospatial evidence fits that edge-coverage proposition better than it fits a pure wilderness-connectivity proposition (aligning with trends we observed in the U.S. previously, where D2D usage skews toward national parks and popular hiking areas).

European geography makes D2D different from Australia, Canada, Chile, and Peru

As mentioned, the U.K.’s early D2D usage profile looks different from lower-density and more geographically expansive markets. World Bank data puts the U.K. at about 283 people per km2 in 2023, compared with about 3.5 in Australia, 4.6 in Canada, 26.5 in Chile, and 26.4 in Peru. The U.K. also has a much smaller land area than those markets and a dense pattern of towns, roads, and transport corridors.

That matters for D2D because the U.K.’s commercial opportunity is not primarily about bringing basic connectivity to vast unserved interiors. Instead, it is about filling residual gaps in a market where terrestrial mobile networks already cover most outdoor premises and much of the country’s landmass, but where rural, coastal, upland, road, indoor, and operator-specific gaps remain highly consequential for users.

This makes O2 Satellite (and future competing D2D services in the U.K. and elsewhere in Europe) strategically different from a terrestrial substitute. O2 says the service lifts its landmass coverage from 89% to 95%, equivalent to an area around two-thirds the size of Wales. But O2 Satellite is not equivalent to Ofcom’s good-quality 4G coverage definition, which includes a sustained 2 Mbps downlink and the ability to sustain a 90-second voice call. O2’s own public help page confirms that standard text messaging and standard voice calls, including emergency calls, are not currently supported on O2 Satellite, and that 999 texts and government emergency alerts are similarly unavailable while connected to the service. This distinction matters commercially (especially for the towerco business model) and politically, since it means satellite coverage is not yet interchangeable with terrestrial mobile coverage such as that delivered by the SRN. That will very likely change as D2D solutions become more capable.

Ofcom’s framework turned U.K. D2D into a first-mover licensable mobile service

The U.K.’s D2D market was the first in Western Europe because the U.K.’s regulator acted first. Ofcom’s December 2025 statement set out a framework for authorizing D2D in mobile spectrum that was the earliest in any European country and broadly followed the Federal Communications Commission’s (FCC) Supplemental Coverage from Space model in the U.S., though it extended authorization across a wider set of eligible bands, including 700 MHz, 800 MHz, 900 MHz, 1400 MHz, 1800 MHz, 2.1 GHz, and 2.6 GHz.

The authorization works through two instruments. The first is a license exemption for handsets and SIM-enabled devices, made under section 8(3) of the Wireless Telegraphy Act 2006, which came into force on February 25, 2026 (applying in the U.K. and territorial seas, excluding the Channel Islands and the Isle of Man). The second is a variation to each mobile network operator’s existing wireless telegraphy license to add a D2D schedule on specific bands. VMO2 received the first such variation on 17 February 2026 on 1800 MHz (Band 3).

This spectrum choice has important operational consequences. Band 3 is a mainstream global LTE and 5G carrier, supported on almost every modern smartphone, which means handset compatibility for O2 Satellite is effectively a question of which devices VMO2 whitelists rather than which devices can physically receive the signal.

The date sequence also shows how directly regulation shaped launch timing. Ofcom received Telefónica (O2 owner) UK’s completed application on January 28, 2026, published notice of its intent to vary the license on February 4, approved the variation on February 12, made the exemption regulations on February 16, and said those regulations were intended to come into effect on February 25. O2 launched O2 Satellite on February 26.

The UK is moving from first launch to competitive testbed

O2 has a first-mover advantage in U.K. D2D, but it is unlikely to remain the only U.K. D2D architecture for long. On April 15, 2026, Ofcom granted a second D2D license variation to VodafoneThree on 900 MHz Band 8, authorizing a service that will run over AST SpaceMobile’s BlueBird satellites through the Satellite Connect Europe joint venture Vodafone Group and AST announced in early 2025. VodafoneThree has said customer trials will begin in summer 2026, with commercial launch targeted for the end of the year.

VodafoneThree is positioning its upcoming D2D service around data, voice, and SMS, which differs from O2’s current app-based data proposition and lack of standard voice and SMS. That reflects a constraint of Starlink’s large first-generation constellation, which is optimised for LTE messaging and low-throughput data, but should improve materially as V2 satellites add more capable payloads, greater cell capacity, and a broader spectrum base to support richer handset services.

The use of the 900 MHz band for VodafoneThree’s service also creates a different radio and device context from O2’s 1800 MHz implementation, although real-world performance will depend on satellite payloads, beam design, power limits, device support, software behavior, and interference constraints as much as on frequency alone.

Taken together, this means that the U.K. could become one of the first markets where two satellite-to-smartphone models are tested under the same national regulator, but with different operator spectrum positions, satellite partners, service propositions, and launch timing.

BT, which runs the U.K.’s largest mobile network through EE, has taken a different path. It announced a Starlink partnership in early 2026, but only for fixed home broadband to hard-to-reach premises, not for D2D handset services. That positions BT as the natural incumbent for any Emergency Services Network satellite D2D overlay, particularly because ESN is designed around resilient nationwide coverage for police, fire, ambulance, and other public safety users. The UK Space Agency opened a formal industry call on this in January 2026, but EE is still left without a consumer D2D handset product at a moment when its two largest mobile competitors are both moving.

The MVNO layer adds a further wrinkle, with Sky Mobile, Tesco Mobile, and Giffgaff on VMO2 and iD Mobile and SMARTY on VodafoneThree, each a potential future D2D reseller.

Europe still has to solve the harmonization question

The UK moved quickly because Ofcom created a national authorization route for D2D in mobile spectrum. The European Union is working through a more complex harmonization problem.

The Radio Spectrum Policy Group’s June 2025 D2D opinion was explicit about that tension. It said that D2D in harmonized ECS mobile bands is currently not possible in EU Member State ECS licenses, because those licenses and technical harmonization decisions were built for terrestrial mobile use. The RSPG recommended that the European Commission issue a mandate to CEPT to develop harmonized technical conditions for D2D-IMT satellite operations in ECS harmonized bands, with follow-up after WRC-27.

The WRC-27 link is important because Agenda Item 1.13 will test the global framework for connecting satellites directly to ordinary IMT handsets through mobile-satellite service allocations, including how those services can coexist with terrestrial mobile networks in bands between 694/698 MHz and 2.7 GHz.

This matters for operators and satellite providers because D2D becomes much more valuable when roaming, interference management, device support, emergency service obligations, lawful intercept, privacy, competition, and market access can be handled consistently across borders. That cross-border layer is now moving from theory into commercial reality, with Rogers extending satellite-to-mobile roaming for Canadian customers in the U.S. and KDDI expanding au Starlink Direct roaming from the U.S. to Canada, the Philippines, and New Zealand from June 2026.

The issue also intersects with the European Commission’s separate assessment of the future use of the EU 2 GHz MSS band, where current authorisations for Viasat and EchoStar expire in May 2027. That band is central to broader debates about 3GPP NTN, MSS continuity, potential new entrants, and European strategic autonomy, but would require new handsets to put into action.

The U.K. is therefore both ahead of much of Europe and still linked to European outcomes. Cross-border coordination, WRC-27 decisions, CEPT technical conditions, satellite market-access safeguards, and the availability of compatible devices will all shape how quickly D2D scales from national firsts to mass-market coverage features.

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.

| April 1, 2026

Alpine Connectivity: How Mobile Networks Perform Across Europe’s Premier Ski Resorts

Austria leads on the slopes, where Alpine-specific auction incentives, public co-funding, and deep low-band spectrum holdings drive superior outcomes in some of Europe’s most challenging yet lucrative radio environments.

Europe’s largest ski resorts are among the most demanding environments for mobile networks. Extreme terrain, steep changes in elevation between sites and users, seasonal demand spikes that can exceed baseline capacity by an order of magnitude, and remote power constraints all combine to make Alpine connectivity an infrastructure challenge that separates well-invested operators from the rest.

For the tens of millions of visitors who pass through the Alps’ major ski resorts each winter, mobile connectivity is no longer a luxury. It underpins safety communications, real-time slope monitoring, social media sharing from the chairlift, and the growing dependence on digital lift passes and resort apps.

To understand how operators are meeting this challenge, we examined Speedtest Intelligence® data across 17 of Europe’s top ski resorts in five Alpine nations (France, Switzerland, Austria, Italy, and Germany) over a 12-month period from March 2025 to March 2026. The analysis draws on consumer-initiated Speedtest results, background signal scans measuring 4G and 5G signal strength and quality, and quality of experience (QoE) metrics including web page load times.

Key Takeaways

  • Austria leads its Alpine peers in median mobile download speeds at top ski resorts. A1 delivered median download speeds of 144.21 Mbps and the strongest 10th percentile outcomes (20.97 Mbps) of all 17 studied operators. The combination of Breitbandstrategie 2030 co-funding, alpine-specific spectrum auction incentives, and deep low-band holdings (Magenta Telekom alone holds 120 MHz across 700 MHz, 800 MHz, and 900 MHz spectrum) has resulted in infrastructure density that sustains performance even in the worst 10% of network conditions.
  • Italy’s mobile network outcomes at top ski resorts are the most competitively balanced of the five countries studied. The country’s four operators sit within just 16.66 Mbps of each other on median download speed (99.06 to 115.72 Mbps), and it is the only market where every operator exceeds 95 Mbps at the median. This reflects the broader national picture in a fiercely competitive and price-sensitive four-player market. The collective 700 MHz spectrum coverage obligation, which requires all operators to jointly reach 99.4% population coverage by June 2027, and the Milano Cortina 2026 Olympic 5G buildout, have driven competition across the Dolomites.
  • France has the widest within-country performance gap at resorts. SFR leads with a median download speed of 165.32 Mbps while Free trails at 48.79 Mbps, a 116.53 Mbps spread that reflects divergent spectrum and technology strategies across French Alpine terrain. While ARCEP’s New Deal Mobile has resulted in 3,200+ shared 4G sites and converging signal strength, shared infrastructure alone has not closed the throughput gap.
  • Germany’s top resorts record the lowest median download speeds (66.43 Mbps) but the best latency and fastest web page loads, while Swisscom leads the pack in Switzerland. Household-centric auction conditions tied to population coverage and transport corridors rather than tourism zones have left Bavarian resorts as a secondary priority in Germany. Without the geographic deployment mandates seen in France or the public co-funding of Austria, operator commercial strategy is the decisive variable, and Salt’s −110 dBm 4G RSRP, the weakest of any operator across all five countries, shows what happens when that incentive is absent.

The Alps poses a unique radio engineering challenge for mobile operators in tough mountain terrain

Providing cellular connectivity at ski resorts in the Alps is an order of magnitude more difficult than serving a comparable number of users in a lowland town. The physics, the logistics, and the economics all work against European operators.

Operators building remote mountain sites increasingly deploy hybrid solar-wind power systems that can supply as much as 80% of daily load from renewables, reducing diesel generator dependence by 70% to 80%, but the upfront engineering cost for a hardened Alpine installation can run 2.5 to 3.5 times that of an equivalent lowland site.

At 3,000 meters, base station equipment needs to operate in temperature ranges of −30 to +40 degrees Celsius across a single year. At −20 degrees Celsius, conventional lithium-ion batteries used for power backup retain roughly 50% of rated capacity, and charging rates must be throttled to prevent damage, meaning that a site designed for eight hours of backup power in summer may deliver fewer than four in winter.

Wet snow, which behaves like rain for signal propagation purposes, produces measurable attenuation on mid-band frequencies, while ice accumulation of even a quarter-inch on antenna radomes “changes the entire deployment scenario” according to engineering guidance. Avalanche protection adds further cost. Steel and concrete snow sheds, flexible net barriers, and reinforced catching dams are standard protective infrastructure around mountain base stations, and in extreme cases operators rely on remote avalanche control systems to manage hazards without placing personnel at risk.

Backhaul is a major constraint. Fiber installation in Alpine terrain requires specialized ASU (Aerial Self-Supporting) cable rated for heavy snowfall and high wind, or trenching through rocky substrate that in many areas can only be excavated during summer months. Where fiber is uneconomical, satellite backhaul can be installed in one to two days at roughly a third of the cost of a microwave link, and A1 in Austria has adopted this approach to connect alpine huts and remote mountain refuges through a partnership with the Austrian and German Alpine associations.

Similarly, Swisscom’s experience in the Ueschinen Valley in the Bernese Oberland illustrates the tradeoffs. When fiber was assessed for this mountain valley serving primarily summer-inhabited alpine huts, the cost-benefit case failed. Instead, Swisscom deployed a 5G site in autumn 2021, providing the entire valley with broadband at a fraction of the fiber cost (and making use of pre-existing fiber for the site backhaul only rather than needing to deploy it to every home).

Equipment logistics present their own challenge. Helicopter delivery to high-altitude sites is standard practice across the Alps, but thin air at elevation reduces rotor lift capacity, and flight windows are constrained by weather. Deutsche Telekom demonstrated one alternative at the Jizerska 50 cross-country ski race in the Czech Jizera Mountains, deploying a drone-mounted base station at 2.3 kilometers altitude that provided connectivity to over 4,400 race participants. This solution provided measured speeds of 95 Mbps download and 34 Mbps upload, a proof-of-concept for temporary surge capacity at mountain events (noting though that drone usage is still at an early stage in the Alpine context and features severe constraints like short battery runtime).

Austria’s sets the Alpine benchmark

Austria emerges as the strongest overall performer in ski resort connectivity in our analysis, combining the highest median download speeds with solid upload performance and latency. The country-level median download of 110.48 Mbps and upload of 17.69 Mbps both rank first among the five countries studied.

The four resorts studied (Ski Arlberg, Saalbach-Hinterglemm-Leogang-Fieberbrunn, Ischgl, and KitzSki) include Austria’s highest-profile ski domains, and our data suggests these destinations benefit from deliberate investment in Alpine connectivity.

Austria Leads European Ski Resort Download Speeds
Speedtest Intelligence® | Mar 2025 – Mar 2026

A1 drives much of this result. At 144.21 Mbps median download and 24.64 Mbps upload, A1 ranks second among all 17 operators on download and first on upload. Critically, A1 also delivers the strongest 10th percentile download performance of any operator at 20.97 Mbps, meaning that even users experiencing the worst 10% of conditions still receive a usable broadband-grade connection.

The gap between A1’s median and 10th percentile download across top Austrian resorts (a ratio of roughly 7:1) is notably tighter than operators like Orange in France (24:1) or Sunrise in Switzerland (21:1), pointing to more consistent infrastructure density across Austrian resort terrain.

Drei (3) follows at 100.89 Mbps median download, and Magenta Telekom sits at 86.85 Mbps. The 57.36 Mbps spread between A1 and Magenta is material but moderate by Alpine standards, and all three operators deliver acceptable performance for typical resort usage.

A1 and SFR Top the Operator Rankings at European Ski Resorts
Speedtest Intelligence® | Mar 2025 – Mar 2026

Analysis of signal strength data further supports A1’s lead. Its median 4G signal strength (known as Reference Signal Received Power or RSRP) of −100 dBm is the strongest of any operator in our data, and its 4G signal quality (RSRQ) of −9 dB is among the best. Magenta Telekom’s 5G signal stands out with a median RSRP of −95 dBm, the strongest 5G reading across all operators, likely reflecting the usage of its leading 120MHz-wide low-band allocation (prime spectrum for wide coverage that travels further in Alpine terrain) across the 700, 800, and 900 MHz bands.

Median web page load times at the top resorts in Austria cluster around 1.4 seconds across all three operators. The fact that A1’s outcomes here are essentially equivalent to Drei despite its significantly higher download speeds illustrate the diminishing returns of raw speed on QoE metrics that are dominated by deeper factors like core interconnect, peering, and content delivery network (CDN) routing. These are the behind-the-scenes infrastructure elements that determine how efficiently data travels between servers and end users, rather than simply how fast the last mile connection can go.

Italy’s tight operator spread points to competitive balance

Italy’s ski resorts deliver the second-highest median download speeds (101.85 Mbps) and present the most balanced competitive picture of any country in the analysis. The three Italian resorts studied (Sella Ronda, Cortina d’Ampezzo, and Kronplatz) sit directly in the zone that received Olympic infrastructure investment, and the post-Games persistence of this network will be a critical test for long-term connectivity improvement in the Dolomites.

All four operators measured (Vodafone, WINDTRE, Iliad, and TIM) sit between 99.06 Mbps and 115.72 Mbps on median download, a total spread of just 16.66 Mbps or roughly 14% variance. The strong showing is somewhat surprising, as it runs counter to Italy’s broader country-level mobile standing. In the latest edition of the Speedtest Global Index™, Italy ranked 52nd globally and ahead of only Germany among the peer group.

A1 and TIM Lead on 4G Signal Strength Across Alpine Resorts
Speedtest Intelligence® | Mar 2025 – Mar 2026 | Median 4G RSRP in dBm (closer to 0 = stronger signal)

Vodafone leads with a median download speed of 115.72 Mbps and with the lowest latency in the country (43 milliseconds). TIM, despite ranking last among Italian operators on median download speed at 99.06 Mbps, delivers the strongest 10th percentile download (12.51 Mbps) and the best 4G RSRP (-100 dBm), consistent with its position as the operator with the largest physical footprint in Italy. TIM’s extensive 4G population coverage and 60 MHz of low-band spectrum give it deep reach that translates into more consistent mountain coverage, even if peak throughput is modestly lower.

WINDTRE deserves attention for upload performance. With a median of 17.98 Mbps, it delivers the highest upload speed of any Italian operator and one of the highest in the full five-country dataset, a finding that may reflect its 170 MHz of mid-band spectrum and capacity optimization choices.

Iliad, the newest entrant, delivers a competitive 101.40 Mbps median download speed, demonstrating that its national roaming and infrastructure-sharing arrangements provide reasonable resort coverage.

Italy features the poorest QoE outcomes among the five studied countries, with median web page load times ranging from 1.4 seconds (Vodafone) to 1.5 seconds (Iliad), roughly 0.1 to 0.2 seconds slower than the fastest country, Germany. This disconnect between relatively strong download speeds and slower page loads strongly suggests that network routing, CDN proximity, and peering arrangements, rather than raw throughput, may be the constraint on user experience in Italian resort areas.

France features the widest operator divide in the Alps

The four French resorts studied (Les 3 Vallees, Paradiski, Tignes-Val d’Isere, and Chamonix-Mont-Blanc) include the largest ski domains in the Alps and represent peak-demand environments where infrastructure sharing alone may not fully address congestion.

France’s overall median download speed of 83.44 Mbps across its top resorts place it fourth among the five studied countries, but this average masks a striking divergence between operators. SFR leads at 165.32 Mbps, the single highest median download of any operator across all five markets. Bouygues Telecom follows at 120.08 Mbps, while Orange sits at 61.62 Mbps and Free trails at 48.79 Mbps. The 116.53 Mbps spread between SFR and Free represents a 70% variance, the widest of any country in the study. It is also notable that this runs counter to the broader national pattern over recent quarters, where Orange leads on speed and consistency ahead of Bouygues, with both well ahead of SFR and Free.

France's Upload Deficit at Ski Resorts Is Pronounced
Speedtest Intelligence® | Mar 2025 – Mar 2026

The French upload story at ski resorts is even more notable. With an overall median upload speed of 6.20 Mbps, France’s upload speed is less than half the next-lowest country (Italy at 13.85 Mbps) and barely a third of Austria’s 17.69 Mbps. Even SFR, France’s fastest downlink operator, manages only 8.38 Mbps upload. Orange’s 10th percentile upload speed drops to 0.35 Mbps, and Free’s to 0.41 Mbps, levels that would render video calls and cloud uploads essentially nonfunctional. This upload deficit may reflect TDD configuration choices on mid-band spectrum, uplink resource allocation policies, or (most likely) backhaul constraints and seasonal congestion specific to French mountain infrastructure.

French Operators Show Extreme Download/Upload Imbalance at Ski Resorts
Speedtest Intelligence® | Mar 2025 – Mar 2026 | Ratio of median download to upload speed (lower = more balanced)

Deeper analysis of the 10th percentile outcomes expose an Alpine consistency problem across all French operators. Orange’s 10th percentile download is just 2.59 Mbps, meaning the worst 10% of user experiences deliver less than 3 Mbps despite a median of 61.62 Mbps. This 24:1 ratio between median and 10th percentile is the highest in our data and points to potential severe congestion or coverage holes within the resort footprint. Free (16:1 ratio) and Bouygues (20:1) show similar patterns. SFR, despite leading on median download, drops to 9.48 Mbps at 10th percentile (17:1) ratio.

Median multi-sever latency performance in France (56 to 58 ms across all four operators) is tightly clustered, suggesting that latency is not the differentiating factor between French operators at ski resorts. The more impactful divergence is on signal. Three of four operators (Orange, SFR, and Bouygues Telecom) feature an identical 4G RSRP of −103 dBm, while Free registers −106 dBm, consistent with Free’s smaller national site footprint and narrower low-band allocation (37 MHz total vs. 47 MHz to 57 MHz for its competitors).

Swisscom dominates, but Switzerland’s overall position is middle of the Alpine pack

Switzerland’s overall median download of 84.76 Mbps at resorts places it third, just above France, despite Swisscom delivering 130.40 Mbps, a figure that would rank among the top operators in any country. The gap reflects the sharp dropoff below Swisscom. Salt records 69.39 Mbps and Sunrise 57.13 Mbps, a 73.27 Mbps spread between top and bottom that represents a wide 56% variance.

On upload and latency, Switzerland is more competitive. Median upload speeds reach 15.91 Mbps overall (second only to Austria), and the median multi-server latency of 48 ms is the second-best, with Swisscom delivering the lowest single-operator latency in our entire dataset at 33 ms.

Analysis of the 10th percentile outcomes highlights a consistency challenge. Sunrise’s 10th percentile download speed is just 2.71 Mbps (a 21:1 ratio to its median) and Salt’s is 4.93 Mbps (14:1). Even Swisscom’s 10th percentile of 8.08 Mbps represents a 16:1 ratio. These floor-performance readings suggest that at peak times or in terrain-challenged areas of Zermatt, Verbier, and the Jungfrau Region, users on any operator can experience severe performance degradation (over and above what is observed in countries like Austria).

Floor Performance Varies Dramatically Across Alpine Operators
Speedtest Intelligence® | Mar 2025 – Mar 2026

Signal data provides a partial explanation for these challenges. Swisscom’s median 4G RSRP of −102 dBm is acceptable but not exceptional. Salt’s median −110 dBm is the weakest 4G reading of any operator across all five countries, consistent with coverage limitations in peripheral Alpine valleys. Salt’s 5G signal at −106 dBm tracks similarly weak.

Swisscom holds the largest low-band allocation in Switzerland (80 MHz across the 700, 800, and 900 MHz bands) and the deepest total spectrum portfolio at 454 MHz across all bands, vs. Sunrise at 294 MHz and Salt at 270 MHz. This spectrum depth, and the operator’s status as the universal service licence holder, likely underpins Swisscom’s clear Alpine advantage.

Switzerland’s market-driven regulatory model, which lacks the kind of direct government co-funding seen in France (ARCEP New Deal Mobile) or Austria (Alpine Infrastructure Fund), places the burden of mountain investment on operator economics.

Germany trades raw speed for latency and QoE

Germany’s ski resort performance is unique. With a 66.43 Mbps median download speed, it ranks last among the five countries. But with a median latency of 42 ms, it delivers the best multi-server responsiveness, 6 ms faster than second-placed Switzerland and 26 ms better than France. Garmisch-Partenkirchen, Oberstdorf, and Sudelfeld are located in Bavaria but do not sit on priority transport corridors, creating a gap where commercial incentive alone drives investment.

Telekom dominates the German market on speed at resorts, recording a 120.58 Mbps median download, more than double its nearest domestic competitor (Vodafone at 58.95 Mbps, O2 at 54.39 Mbps). Telekom also leads on upload (20.22 Mbps) and latency (35 ms), the latter being the second-lowest single-operator figure in our entire dataset after Swisscom.

But Telekom’s 10th percentile download drops to 6.14 Mbps (a 20:1 ratio), notably weaker than the consistency levels seen in Austria. Vodafone’s 10th percentile download speed of 9.38 Mbps actually represents a tighter ratio (6:1), suggesting more even if lower-ceiling coverage. O2’s 5.69 Mbps 10th percentile and modest signal readings (4G RSRP −105 dBm, 5G −103 dBm) likely reflect some rural coverage gaps.

Germany Leads on Latency and Web Page Load Times at Ski Resorts
Speedtest Intelligence® | Mar 2025 – Mar 2026

The profile of results align closely with the relative spectrum position of each German operator. Telekom, with 70 MHz of low-band spectrum (the largest low-band allocation among German operators, spanning 700, 800, and 900 MHz), has the propagation advantage needed for mountain terrain. Vodafone and O2, with weaker low-band positions, therefore compete less effectively at altitude.

Overall, Germany delivers the best median web page load times of any country. Telekom records 1.2 seconds, the lowest figure across all 17 operators, while Vodafone and O2 also perform well at 1.3 seconds each. This QoE advantage aligns with Germany’s latency leadership (and the inherent advantage that it features the highest density of hyperscale infrastructure in the DACH region) and suggests favorable CDN positioning, routing decisions and peering arrangements for German networks.

Policy approaches to Alpine coverage vary widely from subsidized infrastructure sharing to targeted rollout obligations

The regulatory frameworks shaping Alpine connectivity differ materially across the five countries studied, and these differences can help to explain the performance patterns in our data.

France’s policy approach is the most prescriptive. ARCEP’s New Deal Mobile program, announced in January 2018, replaced the traditional auction logic with a commitment-based framework in which operators accepted binding coverage obligations in exchange for administrative renewal of their spectrum rights. The targeted coverage mechanism requires each operator to cover up to 5,000 areas, with government orders issued at a rate of roughly 600 to 800 areas per year per operator and each designated location to be activated within at most 24 months.

Infrastructure sharing is central, but more selectively than the original text implied. Of those 5,000 areas, 2,000 are explicitly intended for four-operator RAN sharing in places where no operator provides “good coverage”; in the remaining areas operators must at least share passive infrastructure, and in some cases active sharing also applies. Separately, operators must reach 99.8% population “good coverage” for voice and SMS, with deadlines staggered between 2028 and 2031 depending on operator. Compliance is enforced with fines. Historically, SFR was penalized €380,000 for failing to cover 47 town centers by the January 2016 deadline, and Orange received a €27,000 penalty for missing five. Critically for ski areas, the 3.4 to 3.8 GHz obligations require 25% of sites in the final two rollout phases to be located in sparsely populated areas.

Austria, meanwhile, has combined spectrum policy with direct public funding. The Breitbandstrategie 2030 targets nationwide symmetric gigabit-capable connectivity by 2030, and the federal government has made €1.4 billion (US $1.6 billion) available through 2026 under Broadband Austria 2030. Coverage obligations from the 2020 multi-band auction (700, 1500, and 2100 MHz) require A1, Hutchison, and T-Mobile Austria to cover 1,702 of 2,100 underserved cadastral communities, roughly 81%, with first deadlines in summer 2022 and most remaining obligations falling in late 2023 and late 2025. Embedding rural buildout directly into the award process helps explain why operators with stronger rural network positions perform relatively well in Alpine terrain.

Five distinct regulatory approaches shape Alpine connectivity investment decisions across Europe.

By contrast, Switzerland takes the most market-driven approach. Swisscom is the universal service licence holder for 2024 to 2031, but the obligation remains modest: basic telephony plus internet access at 10/1 Mbit/s, or 80/8 Mbit/s on request, with reduced rates permitted in exceptional cases. The February 2019 spectrum auction raised around €414 million (US$477 million) and imposed mainly population-based obligations: licensees with 700 MHz spectrum had to reach at least 50% of the population with their own infrastructure by December 2024, while those without 700 MHz spectrum faced a 25% threshold. Those benchmarks can be met without specifically targeting remote Alpine terrain.

Italy’s 700 MHz auction, concluded in October 2018, included a distinctive collective coverage obligation. The 700 MHz licensees must jointly reach 99.4% of the population within 54 months of the band’s July 2022 availability. This joint structure creates a cooperative incentive, since any single operator’s shortfall affects the group, and it is consistent with infrastructure-sharing approaches. Earlier this year, TIM and Fastweb+Vodafone first announced a preliminary RAN-sharing agreement focused on municipalities with fewer than 35,000 inhabitants, and later announced a non-binding initiative to develop up to 6,000 new towers.

Italy’s Piano Italia 5G program provides major public support for fiber backhaul to more than 10,000 existing mobile sites and for new 5G sites in underserved areas, with public funding covering up to 90% of project cost. The Milano Cortina 2026 Winter Olympics added a further layer, with TIM as Official Telecommunications Partner and FiberCop as Fiber Infrastructure Partner connecting venues to high-capacity fiber infrastructure.

Germany’s regulatory model is the most explicitly focused on household and transport coverage. BNetzA’s 2019 auction required at least 100 Mbps for 98% of households in each federal state by the end of 2022, alongside obligations covering motorways, major federal roads, and rail routes. According to operators’ submissions, all three incumbents met the household threshold, but BNetzA said gaps remained in some transport locations and tunnels. That structure is aimed at population density and corridors rather than tourism zones, leaving mountain coverage more dependent on commercial incentive.

Federal support exists, but execution has been slower than the headline ambition suggests. The Mobilfunkstrategie earmarked about €1.1 billion (US$1.3 billion) from the Special Fund for Digital Infrastructure to support up to 5,000 additional masts, and by the end of 2024 the Mobilfunkinfrastrukturgesellschaft had funded 267 sites, with the first masts in operation and the remainder still in the realization phase. Updated obligations adopted in 2025 require at least 50 Mbps over 99.5% of Germany’s land area from 2030, so German ski resorts are still likely to rely primarily on operator-led investment for the foreseeable future.

Bars on the piste matter for competitive differentiation

For operators, the Alpine corridor is both a technical challenge and a strategic opportunity, a place where network quality is highly visible, directly experienced by affluent and digitally engaged visitors, and increasingly essential to resort operations.

The performance landscape across Europe’s top ski resorts reveals a set of structural themes that extend beyond the mountains. Markets where regulation explicitly targets geographic coverage (e.g., France’s ARCEP New Deal Mobile, Austria’s Alpine investment incentives, Italy’s joint coverage obligations) show stronger outcomes than markets where obligations are tied primarily to population thresholds and transport corridors. Germany’s household-centric auction conditions, despite generating significant auction revenue, leave tourism-dependent mountain zones as a secondary priority.

Operator strategy matters as much as regulation. A1 in Austria and Swisscom in Switzerland have built measurable Alpine advantages that function as competitive differentiation. In markets where operator performance is more tightly clustered (Italy) or where infrastructure sharing dominates (France), the quality of the user experience can become more uniform or constrained by shared bottlenecks.

Emerging direct-to-device (D2D) satellite services from providers like SpaceX’s Starlink and AST SpaceMobile represent a potential complementary layer for the highest-altitude and most remote Alpine terrain where terrestrial economics remain prohibitive. Switzerland’s Salt, for example, was the the first operator in Europe to report a successful (albeit non-commercial) Starlink direct-to-cell test, sending satellite-based text messages to a standard 4G smartphone over its mobile spectrum, touting it as a future coverage extension and resilience layer in the most remote areas.

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.

| March 30, 2026

France’s Rail Connectivity Gap: Why Coverage Targets Alone Cannot Close the Mobile Experience Divide on Track

French/Français

Analysis of Speedtest data across 20 major French rail corridors reveals sharp operator disparities in throughput, latency, and quality of experience.

France operates one of Europe’s most heavily used passenger rail networks, carrying hundreds of millions of riders each year across a system that spans high-speed LGV corridors, intercity Intercités routes, and dense regional TER services. As mobile connectivity has shifted from a convenience to a baseline expectation for rail passengers, the quality of cellular service along these corridors has become an infrastructure question in its own right.

The French government and France’s telecom regulator, ARCEP, recognized this early: the 2018 New Deal Mobile established explicit obligations for 4G coverage along approximately 23,000 km of regional rail track, with a target of 90% coverage of daily train services by the end of 2025. By ARCEP’s own reporting, trackside 4G coverage now reaches 97.7% to 99.3% of daily train services, depending on the operator.

Yet coverage presence and coverage quality are not the same thing, and the gap between them is where the passenger experience is actually shaped. Analysis of Speedtest Intelligence® data across a sample of 20 high-traffic French rail routes, encompassing LGV, intercity, and regional corridors, reveals that the passenger’s designated operator matters enormously for throughput, latency, and real-time application performance. The underlying driver is not a mystery: it maps closely to each operator’s spectrum position, particularly in the sub-1 GHz and mid-band ranges most relevant to rail propagation, and network footprint.

This analysis draws on Speedtest Intelligence data collected between March 2025 and March 2026, alongside quality of experience (QoE) and signal metrics, for all four French mobile network operators: Orange, SFR, Bouygues Telecom, and Free. Tests were captured within a 100-meter buffer of the 20 sampled rail corridors.

Key Takeaways:

  • Orange leads with a median download speed of 283.4 Mbps across the sampled rail corridors, 52% faster than second-place SFR (186.5 Mbps) and more than double Free’s 120.4 Mbps. Orange holds the largest sub-1 GHz spectrum portfolio in France at 57.4 MHz, including both 700 and 800 MHz bands, giving it materially deeper low-band reach in the radio environment along rail corridors where propagation and carriage penetration advantages are most pronounced.
  • Multiserver latency splits the market into two distinct tiers: Orange (33 ms) and Bouygues Telecom (34 ms) cluster within a millisecond of each other, while SFR (43 ms) and Free (64 ms) trail significantly. This two-tier pattern persists almost identically across content delivery networks (CDN), gaming, and video conferencing latency, suggesting potential structural network architecture differences in core routing rather than route-specific variation.
  • Orange recorded a leading median 4G signal quality (RSRQ) of -9 dB, a 3 dB advantage over all three rivals (each at -12 dB), despite near-identical signal strength (RSRP) readings of -100 to -104 dBm across operators. The RSRQ gap points to better spectral isolation or more effective load management across Orange’s rail-adjacent cell sites, potentially supported by its 10 MHz mid-band spectrum advantage at 2600 MHz and greater carrier aggregation depth.
  • Application-layer quality of experience (QoE) metrics partially compress the operator gap: median web page load times span only ~0.1 seconds from Orange (1.1 seconds) to Bouygues Telecom (1.2 seconds), and video conferencing jitter varies by just 2 ms across all four operators (4 to 6 ms). However, video start time inverts the throughput ranking, with SFR leading at 1.3 seconds and Bouygues trailing at 1.6 seconds, pointing to differences in CDN peering, edge caching, or video optimization strategy.
  • France’s New Deal Mobile already provides a relatively robust coverage obligation framework. However, ARCEP’s February 2025 enforcement notices to all four operators cites over 300 blocked or delayed deployment sites. This highlights that meeting even geographic coverage targets remains a challenge before quality of service (QoS) metrics can enter the regulatory conversation. Among major European markets, only Germany has moved toward mandating performance floors on rail, while the UK, Spain, and Italy lag further behind.

Orange’s speed lead maps to spectrum depth, not just signal reach

During the period of analysis between March 2025 and March 2026, the download speed disparity observed across French operators on rail is striking. Orange’s median of 283.4 Mbps is approximately 52% above SFR’s 186.5 Mbps, ~110% above Bouygues Telecom’s 135.0 Mbps, and ~135% above Free’s 120.4 Mbps. This is not a marginal gap: it represents a fundamentally different user experience in bandwidth-intensive applications such as video streaming, large file transfers, and cloud-based workflows.

Orange leads on download speed, while Bouygues is ahead on upload
Speedtest Intelligence® | March 2025 – March 2026

Analysis of spectrum data published by GSMA Intelligence provides an explanation for this disparity. Orange holds 257 MHz of total assigned spectrum nationally, the largest portfolio among the four operators, compared with 227 MHz for SFR, 217 MHz for Bouygues Telecom, and 207 MHz for Free. More critically for rail environments, where low-frequency propagation and in-vehicle penetration matter most, Orange leads in sub-1 GHz holdings at 57.4 MHz spanning both the 700 and 800 MHz bands.

SFR and Bouygues Telecom each hold 47.4 MHz of sub-1 GHz spectrum, while Free holds only 37.4 MHz and notably lacks any 800 MHz assignment entirely, relying on 700 MHz and 900 MHz for its low-band coverage layer. Free’s absence from the 800 MHz band, the workhorse of 4G coverage in rural and semi-rural terrain, is a constraint for rail corridor performance.

Orange also holds a 10 MHz mid-band advantage at 2600 MHz (40 MHz vs. 30 MHz for SFR and Bouygues Telecom), which, when combined with its low-band depth, affords greater carrier aggregation flexibility across the frequency layers most relevant to rail. At 3.5 GHz, where Orange holds 90 MHz, the performance impact on rail is limited: the propagation characteristics of C-band are less well suited to the extended inter-site distances and in-carriage penetration losses typical of rail environments.

Orange's unique low-band depth lends it a coverage advantage
Analysis of GSMA Intelligence Data | 2026

Upload speeds tell a different story. Bouygues Telecom leads at 24.7 Mbps, narrowly ahead of Orange at 23.6 Mbps, with SFR at 16.6 Mbps and Free at 9.2 Mbps. The Bouygues-Orange convergence on upload, despite Orange’s clear download lead, may reflect uplink scheduling optimization or time division duplexing (TDD) configuration choices that weight differently across operators.

Analysis of the signal environment confirms this spectrum narrative. Median 4G reference signal received power (RSRP) readings, which measure the strength of the signal from the cell tower, are tightly clustered across operators, ranging from -100 dBm (Bouygues Telecom) to -104 dBm (Free), indicating that all four operators reach rail corridors at broadly comparable signal strength. Yet Orange’s reference signal received quality (RSRQ), which measures the quality of the signal, of -9 dB is 3 dB better than every rival (all at -12 dB).

Since RSRQ captures signal quality relative to total received power including interference, this gap suggests that Orange achieves better spectral isolation on rail, whether through denser site grids, more effective inter-cell interference management, or the greater carrier aggregation depth that its wider spectrum portfolio likely enables.

Orange's 3 dB signal quality advantage persists despite comparable signal strength
Speedtest Intelligence® | March 2025 – March 2026

When coverage does not equal quality: the QoE picture on French rail

While throughput and latency capture raw network capability, QoE metrics reflect what passengers actually feel when using applications. Here, the operator gap narrows considerably at the application layer, even as it remains wide at the access layer.

Median web page load times span just ~0.1 seconds across operators: from Orange at 1.1 seconds to Bouygues Telecom at 1.2 seconds, with SFR (1.2 seconds) and Free (1.2 seconds) in between. That ~10% spread stands in contrast to the 135% gap in raw download throughput, illustrating how application-layer optimization, CDN placement, and protocol efficiency can partially compensate for underlying network differences. A web page load is shaped by DNS resolution, TLS negotiation, and content rendering, all of which are less sensitive to peak throughput than to latency and connection reliability.

Video start time introduces an inversion: SFR leads at 1.3 seconds, followed by Free at 1.4 seconds, Orange at 1.4 seconds, and Bouygues Telecom at 1.6 seconds. The fact that SFR and Free outperform Orange on video start, despite trailing on throughput, points to potential differences in CDN peering arrangements, edge caching topology, or video player optimization that are distinct from raw radio performance. Video start time is heavily influenced by the initial buffering phase, where server proximity and connection setup overhead can outweigh sustained bandwidth.

Application-layer QoE compresses the operator gap despite wide throughput differences
Speedtest Intelligence® | March 2025 – March 2026

Video conferencing metrics reveal a broadly similar picture across all four networks on rail, with median jitter ranging from just 4 ms (Bouygues Telecom) to 6 ms (Free) and mean packet loss from 2.79% (Orange) to 3.47% (Bouygues Telecom). These are not dramatic spreads. Median video conferencing latency falls into the same two-tier structure as multiserver latency: Orange and SFR at 59 ms, Free at 68 ms, and Bouygues Telecom at 77 ms.

CDN and gaming latency mirror this pattern exactly: Orange and SFR share a 59 ms median, Free sits at 68 ms, and Bouygues Telecom at 77 ms. The consistency of this tiering across multiple latency endpoints suggests a core network or peering architecture difference rather than a radio access variation.

Two-tier latency: Orange & Bouygues lead on multi-server, Orange & SFR on apps
Speedtest Intelligence® | March 2025 – March 2026

France’s rail coverage framework: obligations, enforcement, and the quality blind spot

France’s approach to mobile coverage on rail rests primarily on the New Deal Mobile, the landmark 2018 agreement between the government, ARCEP, and all four operators that embedded legally binding coverage commitments into operator frequency licenses. For rail specifically, the framework mandated 4G coverage along 90% of daily train services across approximately 23,000 km of regional rail track by December 31, 2025, with phased obligations for in-vehicle coverage on the 700 MHz band extending to 2030.

ARCEP enforces these obligations through a combination of operator-reported coverage maps, field measurement campaigns exceeding one million data points annually, and its public Mon Reseau Mobile platform. The framework has delivered measurable progress: white zones with zero mobile coverage have fallen from 11% of the territory in 2017 to under 2% (by Q3 2023), and trackside 4G coverage rates now exceed 97% for all operators.

However, ARCEP’s 2024 quality of service campaign found that web page loads succeeded in only around 70% of attempts on average across TGV, Intercites, and TER services, with per-operator success rates varying from around 64% to 79%. Coverage presence, in other words, does not guarantee usable service.

The enforcement reality is challenging. France has demonstrated willingness to levy penalties, but the clearest recent example is from the fixed side rather than mobile: ARCEP fined Orange €26 million (US$30 million) in November 2023 for failing to meet its legally binding FTTH rollout commitments in AMII areas. On the mobile side, ARCEP has also issued multiple formal notices under the New Deal Mobile framework.

Looking ahead, the transition from GSM-R to FRMCS (Future Railway Mobile Communication System), the 5G-based European standard for railway operational communications, will add a new dimension to rail connectivity.

SNCF Réseau appears to be pursuing a hybrid FRMCS model in which dedicated railway infrastructure remains central on the core network, while commercial mobile networks may be used selectively to extend coverage or reduce deployment cost on certain regional or cross-border sections. This will tie commercial network quality on rail directly to operational railway communications for the first time, potentially raising the stakes for on-rail mobile performance beyond passenger experience.

How France’s approach compares: regulatory lessons from Germany, the UK, Spain, and Italy

France sits in the middle of a wide European spectrum on rail mobile regulation, a position that becomes clearer when compared against its four largest peer markets.

Germany has moved furthest toward regulating quality rather than just coverage on rail. Under conditions attached to its 2019 5G spectrum auction, BNetzA set explicit bandwidth floors: 100 Mbps along major railway lines (Hauptschienenwege) and 50 Mbps along other railway lines. Operators have equipped approximately 400 rail tunnels with mobile coverage as part of broader transport corridor obligations. The GINT program has allocated €6.4 million to test 5G feasibility on rail, and FRMCS pilots are expected from 2026. Germany’s approach represents a regulatory philosophy fundamentally different from France’s: it targets what the network delivers, not merely where it reaches.

The United Kingdom sits at the other end of the spectrum. Ofcom’s last dedicated rail connectivity study dates to 2019, and Parliament has repeatedly called for annual reporting that has not materialized. The UK lacks rail-specific spectrum obligations, and responsibility for rail connectivity is fragmented across multiple government departments. The Shared Rural Network targets rural coverage broadly but does not address rail corridors specifically. A Network Rail and Neos Networks infrastructure agreement signals momentum, but a coordinated rail connectivity program is not expected to deliver results before 2027 at the earliest.

Spain has adopted a public-private partnership model. ADIF, the national rail infrastructure manager, signed a €25.5 million (US$29.4 million) contract with Vodafone and SEMI for 5G deployment on high-speed AVE routes, funded in part through the EU Recovery and Resilience Facility. The Spanish approach is project-driven rather than obligation-based, delivering targeted improvements on flagship routes without establishing a universal framework.

Italy has focused on nodes rather than links. FS Group and TIM have partnered on tunnel coverage across high-speed corridors, while INWIT has deployed 5G infrastructure at major stations including Roma Termini. Italy’s PNRR-funded feasibility studies have explored corridor-level connectivity, but AGCOM has not imposed rail-specific coverage or quality obligations. The emphasis remains on ensuring connectivity at stations rather than along the routes between them.

At the EU level, the Connecting Europe Facility (CEF) Digital program allocates approximately €300 million (US$345 million) for 5G corridors along Trans-European Transport Network (TEN-T) routes through 2027. Several France-relevant inception studies have been approved, including projects for the Paris-Brussels and Frejus cross-border rail corridors. The revised TEN-T Regulation (2024/1679) emphasizes digital connectivity as a component of transport infrastructure, but defers specific mandates to member states.

Coverage is a floor, not a ceiling, on rail

France has built one of Europe’s most progressive mobile coverage obligation frameworks for rail, and it has largely eliminated coverage dead zones across the national network thanks to proactive collaboration with industry. Our data reveals that the challenge has now shifted to deeper network optimization, which requires going beyond baseline coverage metrics to understand what passengers actually experience on trains when they have a signal.

The constraints of coverage obligations alone in stimulating better outcomes should be taken into account in the absence of other supporting measures. Orange’s dominant speed performance likely maps to its spectrum advantages, a 57.4 MHz sub-1 GHz portfolio and a 10 MHz lead in mid-band holdings, that no coverage obligation can easily replicate for its rivals. Competitive dynamics beyond the mandate may also play a role here.

As FRMCS approaches and CEF Digital projects advance from inception studies toward deployment, the strategic question shifts from whether trains have signals to what that signal can deliver. Germany’s model of regulating bandwidth floors on rail, rather than just coverage existence, offers a forward-looking template. It could be reinforced with additional metrics for video, latency, QoE, etc. For France and the rest of Europe, the next phase of rail connectivity policy will need to grapple not just with where networks reach, but with how well they perform when they get there.


L’écart de connectivité ferroviaire en France : pourquoi les seuls objectifs de couverture ne peuvent pas combler le fossé d’expérience mobile sur les voies ferrées

L’analyse des données Speedtest sur 20 corridors ferroviaires majeurs français révèle des disparités nettes entre opérateurs en termes de débit et de latence, exposant les limites d’un cadre réglementaire qui impose la portée géographique mais pas encore la qualité de service.

La France opère l’un des réseaux ferroviaires de passagers les plus intensément utilisés d’Europe, transportant des centaines de millions de voyageurs chaque année sur un système qui s’étend des corridors LGV à grande vitesse, aux services Intercités et aux services régionaux TER denses. À mesure que la connectivité mobile est passée d’une commodité à une attente de base pour les passagers des trains, la qualité du service cellulaire le long de ces corridors est devenue une question d’infrastructure à part entière.

Le gouvernement français et le régulateur français des télécommunications, l’ARCEP, ont reconnu ce fait précocement : le New Deal Mobile de 2018 a établi des obligations explicites pour la couverture 4G le long d’environ 23 000 km de voies ferrées régionales, avec un objectif de 90 % de couverture des services de trains quotidiens d’ici fin 2025. Selon le propre rapport de l’ARCEP, la couverture 4G en bordure de voies ferrées atteint désormais 97,7 % à 99,3 % des services de trains quotidiens, selon l’opérateur.

Cependant, la présence de couverture et la qualité de la couverture ne sont pas la même chose, et c’est le fossé entre elles qui façonne réellement l’expérience des passagers. L’analyse des données Speedtest Intelligence® sur un échantillon de 20 routes ferroviaires françaises à fort trafic, englobant des corridors LGV, Intercités et régionaux, révèle que l’opérateur auquel s’abonne un passager importe énormément pour le débit, la latence et les performances des applications en temps réel. Le facteur sous-jacent n’est pas un mystère : il correspond étroitement au portefeuille de spectre de chaque opérateur, particulièrement dans les bandes sub-1 GHz et moyennes les plus efficaces pour la propagation ferroviaire, ainsi qu’à son empreinte réseau.

Cette analyse s’appuie sur les données Speedtest Intelligence collectées entre mars 2025 et mars 2026, ainsi que sur les métriques de qualité d’expérience (QoE) et de signal, couvrant les quatre opérateurs mobiles français : Orange, SFR, Bouygues Telecom et Free. Les tests ont été capturés dans un rayon de 100 mètres autour des 20 corridors ferroviaires échantillonnés.

Enseignements clés :

  • Orange domine avec un débit descendant médian de 283,40 Mbps sur les corridors ferroviaires échantillonnés, 52 % plus rapide que SFR en deuxième position (186,53 Mbps) et plus du double des 120,41 Mbps de Free. Orange détient le plus grand portefeuille de spectre sub-1 GHz en France avec 57,4 MHz, incluant les bandes 700 et 800 MHz, lui donnant une portée clairement plus importante en bande basse dans l’environnement radio le long des corridors ferroviaires où les avantages de propagation et de pénétration dans les wagons sont les plus importants.
  • La latence multi-serveurs divise le marché en deux niveaux distincts : Orange (33 ms) et Bouygues Telecom (34 ms) se situent à moins d’une milliseconde l’une de l’autre, tandis que SFR (43 ms) et Free (64 ms) accusent un retard important. Ce schéma à deux niveaux persiste presque identiquement sur les réseaux de distribution de contenu (CDN), les jeux et la conférence vidéo en latence, suggérant des différences potentielles d’architecture réseau structurelle dans l’acheminement central plutôt qu’une variation spécifique aux itinéraires.
  • Orange a enregistré une qualité de signal 4G médiane dominante (RSRQ) de -9 dB, un avantage de 3 dB sur les trois rivaux (chacun à -12 dB), malgré des lectures de puissance de signal (RSRP) pratiquement identiques de -100 à -104 dBm sur tous les opérateurs. L’écart RSRQ pointe vers une meilleure isolation spectrale ou une gestion de charge plus efficace sur les sites cellulaires adjacents aux voies ferrées d’Orange, potentiellement soutenue par son avantage de spectre en mi-bande de 10 MHz sur 2600 MHz et une profondeur d’agrégation de porteuses plus importante.
  • Les métriques de qualité d’expérience (QoE) au niveau des applications compriment partiellement l’écart opérateur : les temps de chargement des pages Web médians s’étendent sur seulement environ 0,1 secondes d’Orange (1,1 secondes) à Bouygues Telecom (1,2 secondes), et la gigue (jitter) de conférence vidéo varie de seulement 2 ms sur les quatre opérateurs (4 à 6 ms). Cependant, le temps de démarrage vidéo inverse le classement du débit, SFR se classant en tête à 1,3 secondes et Bouygues en retard à 1,6 secondes, pointant vers des différences dans l’appairage CDN, l’utilisation de serveur cache en périphérie ou la stratégie d’optimisation vidéo.
  • Le New Deal Mobile français fournit déjà un cadre d’obligation de couverture relativement robuste. Cependant, les avis d’application de février 2025 de l’ARCEP à tous les quatre opérateurs citent plus de 300 sites de déploiement bloqués ou retardés. Cela souligne que respecter même les objectifs de couverture géographique reste un défi avant que les métriques de qualité de service (QoS) puissent entrer dans le débat réglementaire. Parmi les grands marchés européens, seule l’Allemagne a avancé pour mandater des minimums de performance sur les voies ferrées, tandis que le Royaume-Uni, l’Espagne et l’Italie accusent davantage de retard.

L’avance en vitesse d’Orange correspond à la profondeur spectrale, pas seulement à la portée du signal

Au cours de la période d’analyse entre mars 2025 et mars 2026, la disparité de vitesse de téléchargement observée entre les opérateurs français sur les voies ferrées est frappante. La médiane d’Orange de 283,40 Mbps est environ 52 % supérieure aux 186,53 Mbps de SFR, ~110 % supérieure aux 135,02 Mbps de Bouygues Telecom, et ~135 % supérieure aux 120,41 Mbps de Free. Ce n’est pas un écart marginal : il représente une expérience utilisateur fondamentalement différente dans les applications intensives en bande passante telles que la transmission vidéo, les transferts de fichiers volumineux et les applications cloud.

Débits descendants et montants des opérateurs sur les corridors ferroviaires français
Speedtest Intelligence® | mars 2025 – mars 2026

L’analyse des données de spectre publiées par GSMA Intelligence fournit une explication de cette disparité. Orange détient 257 MHz de spectre assigné au total au niveau national, le plus grand portefeuille parmi les quatre opérateurs, comparé avec 227 MHz pour SFR, 217 MHz pour Bouygues Telecom et 207 MHz pour Free. Plus crucialement pour les environnements ferroviaires, où la propagation en basse fréquence et la pénétration dans les wagons comptent le plus, Orange domine les allocations sub-1 GHz avec 57,4 MHz couvrant à la fois les bandes 700 et 800 MHz.

SFR et Bouygues Telecom détiennent chacun 47,4 MHz de spectre sub-1 GHz, tandis que Free n’en détient que 37,4 MHz et ne dispose notamment d’aucune attribution dans la bande 800 MHz, s’appuyant sur 700 et 900 MHz pour sa couche de couverture en bande basse. L’absence de Free de la bande 800 MHz, l’outil de base de la couverture 4G dans le terrain rural et semi-rural, est une contrainte pour la performance des corridors ferroviaires.

Orange détient également un avantage en mi-bande de 10 MHz sur 2600 MHz (40 MHz comparés à 30 MHz pour SFR et Bouygues Telecom), qui, combiné avec sa profondeur en bande basse, lui confère une flexibilité d’agrégation de porteuses plus importante sur les couches de fréquence les plus efficaces pour les voies ferrées. Sur 3,5 GHz, où Orange détient 90 MHz, l’impact sur la performance ferroviaire est limité : les caractéristiques de propagation de la bande C conviennent moins bien aux distances inter-sites étendues et aux pertes de pénétration dans les wagons typiques des environnements ferroviaires.

Portefeuille de spectre des opérateurs mobiles français
Analyse des données GSMA Intelligence | 2026

Les débits montants dénotent une toute autre réalité. Bouygues Telecom mène à 24,75 Mbps, de justesse devant Orange à 23,59 Mbps, avec SFR à 16,59 Mbps et Free à 9,18 Mbps. La convergence Bouygues-Orange sur la vitesse ascendante, malgré la nette avance en descente d’Orange, peut refléter des choix d’optimisation de planification ou de configuration TDD qui pèsent différemment selon les opérateurs.

L’analyse de l’environnement de signal confirme cette observation. Les lectures médianes de puissance du signal de référence reçu 4G (RSRP), qui mesurent la force du signal de la tour cellulaire, sont étroitement regroupées entre opérateurs, allant de -100 dBm (Bouygues Telecom) à -104 dBm (Free), indiquant que les quatre opérateurs atteignent les corridors ferroviaires à une intensité de signal comparable. Cependant, la qualité du signal de référence reçu (RSRQ) d’Orange, qui mesure la qualité du signal, de -9 dB, est 3 dB meilleure que celle de chaque rival (tous à -12 dB).

Étant donné que RSRQ capture la qualité du signal par rapport à la puissance totale reçue incluant les interférences, cet écart suggère qu’Orange réalise une meilleure isolation spectrale sur les voies ferrées, que ce soit par des grilles de sites plus denses, une gestion plus efficace des interférences inter-cellules, ou la profondeur d’agrégation de porteuses plus importante que son portefeuille de spectre plus large favorise probablement.

Qualité du signal 4G sur les corridors ferroviaires français
Speedtest Intelligence® | mars 2025 – mars 2026

Quand la couverture ne rime pas avec qualité : l’état de la QoE sur les lignes ferroviaires françaises

Bien que le débit et la latence capturent la capacité réseau brute, les métriques de qualité d’expérience reflètent ce que les passagers perçoivent réellement lorsqu’ils utilisent les applications. Ici, l’écart opérateur se rétrécit considérablement au niveau application, même s’il reste large au niveau de l’accès.

Les temps de chargement des pages Web médians s’étendent sur seulement environ 0,1 secondes entre opérateurs : d’Orange à 1,1 secondes à Bouygues Telecom à 1,2 secondes, avec SFR (1,2 secondes) et Free (1,2 secondes) entre les deux. Cet écart d’environ 10 % contraste fortement avec l’écart de 135 % en débit de téléchargement brut, illustrant comment l’optimisation au niveau application, le placement CDN et l’efficacité des protocoles peuvent partiellement compenser les différences réseau sous-jacentes. Un chargement de page Web est façonné par la résolution DNS, la négociation TLS et le rendu de contenu, tout cela étant moins sensible au débit maximal qu’à la latence et à la fiabilité de la connexion.

Le temps de démarrage vidéo introduit une inversion du classement : SFR se classe en tête à 1,3 secondes, suivi de Free à 1,4 secondes, Orange à 1,4 secondes et Bouygues Telecom à 1,6 secondes. Le fait que SFR et Free surpassent Orange au démarrage vidéo, malgré un retard de débit, pointe vers des différences potentielles dans les arrangements d’appairage CDN, la topologie du serveur cache en périphérie ou l’optimisation du lecteur vidéo qui sont distinctes de la performance radio brute. Le temps de démarrage vidéo est fortement influencé par la phase de buffering initiale, où la proximité du serveur et la surcharge pour l’établissement de connexion peuvent surpasser la bande passante soutenue.

Métriques de qualité d'expérience sur les voies ferrées françaises
Speedtest Intelligence® | mars 2025 – mars 2026

Les métriques de conférence vidéo révèlent un état largement similaire sur les quatre réseaux sur les voies ferrées, avec une gigue (jitter) médiane variant de seulement 4 ms (Bouygues Telecom) à 6 ms (Free) et une perte de paquets moyenne de 2,79 % (Orange) à 3,47 % (Bouygues Telecom). Ce ne sont pas des écarts dramatiques. La latence de conférence vidéo médiane tombe dans la même structure à deux niveaux que la latence multi-serveurs : Orange et SFR à 59 ms, Free à 68 ms et Bouygues Telecom à 77 ms.

Les latences CDN et jeux reflètent exactement ce modèle : Orange et SFR partagent une médiane de 59 ms, Free se situe à 68 ms et Bouygues Telecom à 77 ms. La cohérence de cette hiérarchisation sur plusieurs points de terminaison de latence suggère une différence d’architecture réseau central ou d’appairage plutôt qu’une variation d’accès radio.

Niveaux de latence des opérateurs français sur les voies ferrées
Speedtest Intelligence® | mars 2025 – mars 2026

Le cadre de couverture ferroviaire français : obligations, application et l’angle mort de la qualité

L’approche française de la couverture mobile sur les voies ferrées repose principalement sur le New Deal Mobile, l’accord majeur de 2018 entre le gouvernement, l’ARCEP et les quatre opérateurs, intégrant des engagements de couverture juridiquement contraignants dans les licences de fréquences des opérateurs. Pour les voies ferrées spécifiquement, le cadre mandate la couverture 4G le long de 90 % des services de trains quotidiens sur environ 23 000 km de voies ferrées régionales d’ici le 31 décembre 2025, avec des obligations échelonnées pour la couverture dans les wagons sur la bande 700 MHz s’étendant à 2030.

L’ARCEP applique ces obligations par une combinaison de cartes de couverture rapportées par les opérateurs, de campagnes de mesure sur le terrain dépassant un million de points de données annuels et de sa plateforme publique Mon Réseau Mobile. Ce dispositif a permis des avancées tangibles : les zones blanches, dépourvues de toute couverture mobile, sont passées de 11 % du territoire en 2017 à moins de 2 % (au T3 2023), et les taux de couverture 4G le long des voies ferrées dépassent désormais les 97 % pour l’ensemble des opérateurs.

Cependant, la campagne de qualité de service de l’ARCEP en 2024 a constaté que les chargements de pages Web n’ont réussi que dans environ 70 % des tentatives en moyenne sur les services TGV, Intercités et TER, avec des taux de réussite par opérateur variant d’environ 64 % à 79 %. La présence de couverture, en d’autres termes, ne garantit pas un service utilisable.

Dans les faits, faire respecter ces obligations s’avère complexe. Si la France a déjà prouvé sa volonté de sévir, l’exemple récent le plus marquant concerne le réseau fixe et non le mobile : en novembre 2023, l’ARCEP a infligé une amende de 26 millions d’euros (30 millions de dollars) à Orange pour le non-respect de ses engagements juridiquement contraignants de déploiement FTTH en zone AMII. Sur le front du mobile, l’ARCEP a également adressé de multiples mises en demeure dans le cadre du New Deal Mobile.

À l’avenir, la transition du GSM-R vers le FRMCS (Future Railway Mobile Communication System), la norme européenne basée sur 5G pour les communications opérationnelles ferroviaires, ajoutera une nouvelle dimension à la connectivité ferroviaire.

SNCF Réseau semble s’orienter vers un modèle FRMCS hybride : l’infrastructure ferroviaire dédiée resterait au cœur du réseau principal, tandis que les réseaux mobiles commerciaux pourraient être mis à contribution de façon ciblée pour étendre la couverture ou réduire les coûts de déploiement sur certains tronçons régionaux ou transfrontaliers. Pour la première fois, la qualité des réseaux commerciaux sur le domaine ferroviaire sera directement corrélée aux communications opérationnelles. Les enjeux liés à la connectivité mobile sur les rails s’en trouveront ainsi décuplés, dépassant largement le simple cadre de l’expérience voyageur.

Comment l’approche française se compare : leçons réglementaires d’Allemagne, du Royaume-Uni, d’Espagne et d’Italie

La France se situe au milieu d’un large spectre européen en matière de régulation mobile ferroviaire, une position qui devient plus claire lorsqu’elle est comparée à ses quatre plus grands marchés pairs.

L’Allemagne a avancé le plus loin vers la régulation de la qualité plutôt que seulement la couverture sur les voies ferrées. Selon les conditions attachées à son enchère de spectre 5G de 2019, BNetzA a fixé explicitement le minimum pour la bande passante : 100 Mbps le long des lignes ferroviaires majeures (Hauptschienenwege) et 50 Mbps le long des autres lignes ferroviaires. Les opérateurs ont équipé environ 400 tunnels ferroviaires avec une couverture mobile dans le cadre d’obligations plus larges de corridors de transport. Le programme GINT a alloué 6,4 millions d’euros pour tester la faisabilité de la 5G sur les voies ferrées, et les pilotes FRMCS sont attendus à partir de 2026. L’approche allemande représente une philosophie réglementaire fondamentalement différente de celle de la France : elle cible ce que le réseau fournit, pas simplement où il atteint.

Le Royaume-Uni se situe à l’autre extrémité du spectre. La dernière étude dédiée d’Ofcom sur la connectivité ferroviaire date de 2019, et le Parlement a appelé à plusieurs reprises à des rapports annuels qui ne se sont pas matérialisés. Le Royaume-Uni n’a pas d’obligations de spectre ferroviaire spécifiques, et la responsabilité de la connectivité ferroviaire est fragmentée entre plusieurs départements gouvernementaux. Le Shared Rural Network cible largement la couverture rurale mais ne s’adresse pas spécifiquement aux corridors ferroviaires. Un accord d’infrastructure entre Network Rail et Neos Networks signale un élan, mais un programme de connectivité ferroviaire coordonné n’est pas attendu pour fournir des résultats avant 2027 au plus tôt.

L’Espagne a adopté un modèle de partenariat public-privé. ADIF, le gestionnaire national des infrastructures ferroviaires, a signé un contrat de 25,5 millions d’euros avec Vodafone et SEMI pour le déploiement de 5G sur les routes AVE à grande vitesse, financé en partie par la Facilité pour la reprise et la résilience de l’UE. L’approche espagnole est basée sur des projets plutôt que sur des obligations, fournissant des améliorations ciblées sur les itinéraires phares sans établir un cadre universel.

L’Italie s’est concentrée sur les nœuds plutôt que sur les liens. Le groupe FS et TIM se sont associés sur la couverture des tunnels sur les corridors à grande vitesse, tandis qu’INWIT a déployé l’infrastructure 5G dans les principales gares incluant Roma Termini. Les études de faisabilité financées par le PNRR de l’Italie ont exploré la connectivité au niveau des corridors, mais l’AGCOM n’a pas imposé d’obligations de couverture ou de qualité ferroviaires. L’accent demeure sur l’assurance de la connectivité aux gares plutôt que le long des itinéraires qui les séparent.

Au niveau de l’UE, le programme Connecting Europe Facility (CEF) Digital alloue environ 300 millions d’euros pour les corridors 5G le long des routes du Réseau transeuropéen de transport (RTE-T) jusqu’à 2027. Plusieurs études d’amorçage pertinentes pour la France ont été approuvées, incluant des projets pour les corridors ferroviaires transfrontaliers Paris-Bruxelles et Fréjus. Le règlement révisé RTE-T (2024/1679) souligne la connectivité numérique comme composante de l’infrastructure de transport, mais renvoie les obligations spécifiques aux États membres.

La couverture est un plancher, pas un plafond, sur les voies ferrées

La France a construit l’un des cadres d’obligations de couverture mobile les plus progressifs d’Europe pour les voies ferrées, et elle a largement éliminé les zones mortes de couverture sur le réseau national grâce à une collaboration proactive avec l’industrie. Nos données révèlent que le défi a maintenant basculé vers une optimisation réseau plus profonde, qui nécessite d’aller au-delà des simples métriques de couverture de base pour comprendre ce que les passagers vivent réellement sur les trains quand ils ont un signal.

En l’absence d’autres mesures de soutien, il convient de prendre en compte les contraintes des obligations de couverture seules dans la stimulation de meilleurs résultats. Par exemple, la performance de débit dominante d’Orange est grâce à son portefeuille sub-1 GHz de 57,4 MHz et son avantage de spectre mi-bande de 10 MHz (et peut également refléter la concurrence entre opérateurs au-delà du mandat), des avantages qu’aucune obligation de couverture ne peut facilement reproduire pour ses rivaux.

À mesure que FRMCS approche et que les projets CEF Digital progressent des études initiales au déploiement, la question stratégique passe de savoir si les trains ont des signaux à ce que ce signal peut fournir. Le modèle allemand de régulation des planchers de bande passante sur les voies ferrées, plutôt que simplement l’existence de la couverture, offre un modèle avant-gardiste. Il pourrait être renforcé par des métriques supplémentaires pour la vidéo, la latence, la QoE, etc. Pour la France et le reste de l’Europe, la prochaine phase de la politique de connectivité ferroviaire devra s’attaquer non seulement à la couverture géographique des réseaux, mais aussi à la performance de ces derniers une fois sur place.

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.

, , , | March 11, 2026

MWC 2026 Wrap-Up: The Analyst Take on Satellite, 6G, Sovereignty, and AI

This year, the conversations at Mobile World Congress (MWC) 2026 in Barcelona shifted dramatically. While previous years focused heavily on intrinsic challenges for the telecom sector—the need for 5G monetization, the untapped enterprise opportunity, the calls for “fair share”, and the need for network consolidation—this year’s show focused more on upside. The show floor and our discussions were dominated by opportunities around topics such as satellite, sovereignty, amidst the emerging age of AI, and with a view towards the arrival of 6G. Following the event, our Ookla Research analysts—Mike Dano, Mark Giles, Luke Kehoe, and Karim  Yaici—sat down to cut through the noise.

The mainstreaming of satellite and NTN

Satellite connectivity and Non-Terrestrial Networks (NTN) have officially moved from a niche talking point to a core architectural consideration.

Key announcements:

  • Starlink’s Next-Gen Push: Starlink held a massive keynote to announce its second-generation satellite constellation for direct-to-device (D2D), slated to begin offering services in 2028. Deutsche Telekom was announced as their first official customer for this new constellation. The introduction of the “Starlink Mobile” brand looks to be an important but still early stepping stone in a journey toward a more fully-fledged mobile service. 
  • The AST SpaceMobile Counter: AST SpaceMobile continues to make a huge amount of noise, bolstered by a major pre-MWC announcement regarding their deepening partnership and joint venture with Vodafone through Satellite Connect Europe

Our take: Low-Earth-orbit (LEO)-based D2D  satellite connectivity is graduating from a novelty feature for hikers into a standard “resilience layer” for mass-market mobile networks. The super-bundle of the future will integrate fiber, cellular, and satellite into a single service that automatically fails over when one link drops. What we are witnessing is a massive race to conquer space.

On one side, you have the operator-backed AST SpaceMobile, which operates without a consumer-facing brand and continues to face delays in constellation buildout. On the other, you have Starlink, which has been incredibly strategic about exposing its brand everywhere,from airline Wi-Fi to fixed broadband, and now mobile. The big question moving forward is whether Starlink’s D2D offering stays at that or is just a stepping stone toward a much more capable, hybrid space-terrestrial mobile offering in the future.

5G-Advanced, 5G Standalone, and the 6G horizon

The industry is balancing the need to monetize existing 5G investments with the architectural groundwork required for 6G.

Key announcements:

  • 5G slice validation: Ookla showcased its own collaboration at the event: an industry-first methodology for testing 5G network slices, co-developed with Ericsson. This specialized proof of concept in the Speedtest app enables real-time validation of differentiated 5G connectivity for ultra-low latency and mission-critical reliability (think slices optimized for gaming or video conferencing).
  • 6G timelines: SoftBank laid down a marker, stating they expect to deliver initial 6G services in 2029, emphasizing the need for massive 400 MHz bands to operate effectively.
  • AI-RAN commercialization: Nokia executives promised commercial AI-RAN deployments (in collaboration with Nvidia) by 2027, bridging the gap between 5G-Advanced and 6G.

Our take: 5G Standalone (SA) was frequently mentioned as a prerequisite and stepping stone to 6G networks. While our data, released just before MWC, shows huge variation in 5G SA adoption globally, it’s clear that leading operators are leaning into the technology, to launch new services and drive competitive advantage.

When it comes to 6G, the technical momentum is real, but it is tempered by economic caution. Many European operators remain hesitant about undertaking another massive capital expenditure so soon. However, the U.S. likely targeting the 2028 Olympics for early pre-commercial 6G deployment creates a global race dynamic, with rival markets and even operators within the U.S., forced to respond.

Digital sovereignty across the stack

The need for secure, localized telecom and cloud infrastructure is set to become a defining procurement criterion for enterprises and the public sector, especially in Europe. There is now a rigid demand for independent, Europe-anchored solutions that remain within local control. This drive for sovereignty isn’t just about satellite; it extends across the entire telecom stack, particularly the cloud.

Key announcements:

  • Deutsche Telekom’s Cloud Ambitions: DT delivered a standout presentation at its booth, detailing its heavy investment in its cloud business. It noted it is currently at roughly 80% feature parity with AWS and is pushing for 100% by year-end, with plans to expand its cloud availability regions beyond the DACH.
  • Orange’s Pan-European Sovereign Edge: Orange joined forces with Deutsche Telekom, Telefónica, TIM, and Vodafone to launch the “European Edge Continuum.” This first-of-its-kind federated edge cloud allows enterprises to deploy applications seamlessly across all five operator networks via a single entry point, providing a secure, sovereign alternative to U.S.-based public hyperscalers.
  • Post-quantum security: Several Tier-1 operators showcased active implementations of Post-Quantum Cryptography (PQC) across their networks, ensuring that sovereign data remains secure against future quantum computing threats.

Our take: Digital sovereignty is driving renewed focus from leading telcos in their B2B operations, as demand rises for local platforms capable of hosting sovereign AI models and evolving cloud workloads. For telcos, this is a massive B2B opportunity. By offering secure, localized solutions that align with national data regulations, operators can position themselves as the active shield of the digital economy, moving far beyond basic connectivity.

AI in telecom: moving beyond an efficiency play

AI was omnipresent at MWC26, but the narrative has evolved from generative AI chatbots to “agentic AI” and network-level intelligence.

Key announcements:

  • AI-Enhanced Calling: Asian operators (like LG Uplus and China Mobile), along with Deutsche Telekom, are injecting new innovation into a historically stagnant area: the calling experience, using AI for real-time translation and network-driven functions like enhanced interactive video. Deutsche Telekom showcased similar examples, drawing on recent announcements around AI-translated calling features in the U.S.  
  • AT&T’s connected AI: AT&T outlined its industrial edge strategy, partnering with major hyperscalers to position its fiber and edge infrastructure as the backbone for enterprise AI workloads.

Our take: AI is framing nearly every technical discussion in telecom, but it’s clear that most of the focus has been on using AI to streamline operations and target cost-cutting. MWC 26 saw this evolve, with developments targeting improvements to the user experience – most notably for voice services, and a renewed focus on the edge with AI-RAN.

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.

| February 25, 2026

Repeated Storms Test Portugal's Network Resilience and Spotlight the Role of Satellite Connectivity

Mobile network download speeds declined by more than 50%, while Starlink usage experienced a nearly 200% increase compared to pre-storm baselines

Less than a year after the April 2025 Iberian Peninsula blackout exposed deep vulnerabilities in Portugal’s telecom infrastructure, the country faced another severe test. Between late January and early February 2026, a rapid succession of powerful extratropical cyclones battered the country, knocking out power to over a million customers, disrupting mobile connectivity for hundreds of thousands, and triggering a dramatic spike in satellite broadband usage. Speedtest Intelligence® data captures the scale of network impact and the emerging role of low-earth orbit (LEO) satellite connectivity as a layer of redundancy when terrestrial networks falter.

Key Takeaways:

  • Median mobile download speeds in Portugal fell by as much as 52.4% from their pre-storm baseline, dropping from 107.3 Mbps to just 51.1 Mbps at their lowest point on February 8, as successive storms compounded network strain.
  • Mobile upload speeds declined by up to 46.6%, while latency increased by 15.6% and jitter by 27.1% during the worst of the disruption, reflecting significant network congestion and infrastructure stress.
  • Starlink user activity in Portugal surged by approximately 196% above pre-storm levels at its peak on February 12, with elevated adoption persisting well into late February even as mobile networks began to stabilize.
  • The data highlights a clear network substitution pattern, with Starlink activity climbing in near-lockstep with mobile speed declines and fixed network disruptions, reinforcing the case for satellite as a meaningful resilience layer during prolonged terrestrial outages.
  • Portugal, like many European countries outside the Nordics, continues to lack any binding requirements for specific minimum backup power levels at mobile sites. Recent policy developments in Switzerland and in the EU’s Digital Networks Act (DNA) suggest resilience planning is moving from concept to practical action.

A devastating storm sequence

Storm Kristin made landfall in Portugal’s Leiria district on the night of January 28, 2026, bringing record-breaking winds of over 200 km/h in the Coimbra region and generating over 1,500 emergency incidents in a single night. The storm was the most destructive to hit the country in recent memory, surpassing wind speed records previously held by Hurricane Leslie. Initial disruption was severe, with Reuters reporting more than 3,000 weather-related incidents and electricity distributor E-REDES indicating that outage levels had earlier reached 855,000 customers before restoration work began to reduce that figure.

On the grid side, the damage split across transmission and distribution layers. REN reported 61 very high-voltage pylons knocked down during Storm Kristin and 774 km of very high-voltage lines out of operation, which it said was equivalent to about 7% of Portugal’s transmission grid. E-REDES separately reported more than 600 damaged medium-voltage poles and said that more than one million customers had been left without power at one stage of the event. This distinction is significant because it highlights how resilience bottlenecks emerge across multiple network tiers, not only at the local distribution level.

Telecom disruption was also prolonged. Paulo Fernandes (head of the Central Region Reconstruction Mission Structure) said the affected area started with 307,900 mobile and landline users without communications on 30 January, and that nearly 84,000 customers in the central region still lacked communications almost three weeks later. He also said around 40% of cases were linked to the restoration of electricity supply to mobile sites.

This power dependency aligns with local expert commentary. INESC TEC-linked analysis highlighted that many telecom outages were driven first by loss of electricity at network sites, with a share of the remainder linked to infrastructure faults such as fiber breaks. It was also reported that ANACOM had recommended activation of national roaming, which could help by allowing users to attach to alternative networks where available.

Mobile network performance degradation was severe and sustained, with recovery still ongoing

Analysis of Speedtest Intelligence data paints a detailed picture of how Portugal’s mobile networks responded to the storm sequence. To assess the impact, we established a pre-storm performance baseline using daily median values from January 3 through January 27, 2026, then measured deviations across three distinct phases of disruption.

Prior to the storms, Portugal’s mobile networks were delivering a median download speed of 107.3 Mbps and median upload speed of 15.7 Mbps, with a multi-server latency of 33.5 ms. These figures are consistent with a well-performing mobile market (ranking in the top 30 globally in the latest iteration of the Speedtest Global Index).

The onset of Storm Kristin on January 28 triggered an immediate and sharp decline. Median download speeds fell to 64.5 Mbps that day, a 39.9% drop from baseline, while upload speeds declined 37.4% to 9.8 Mbps. Latency spiked 15.6% to 39 ms and jitter surged 27.1% to 10 ms, indicating significant network congestion as damaged infrastructure concentrated traffic on surviving cells (likely compounded by the loss of fixed connectivity in homes driving more traffic onto the depleted mobile grid).

Rather than recovering, network performance continued to deteriorate in the days that followed as Storms Leonardo and Marta exacerbated the damage. During the sustained disruption phase from February 1 through 14, average median download speeds fell to 59.9 Mbps, a 44.1% decline from baseline. The single worst day came on February 8, during Storm Marta, when median download speeds bottomed out at just 51.1 Mbps, a 52.4% decline. Upload speeds during this phase averaged just 9.5 Mbps, down 39.1% from baseline.

Latency and jitter, often overlooked but critical indicators of quality of experience (QoE) in interactive applications like video conferencing, told a similar story. Median latency during the sustained phase rose to 37 ms, a 10.5% increase over baseline, while jitter averaged 9 ms, up 21.6%. Elevated jitter in particular can reflect the instability characteristic of a network under duress, where routing paths shift unpredictably as infrastructure comes on and offline.

By mid-to-late February, partial recovery was underway. Median download speeds during the February 15 through February 23 period rose to 69.8 Mbps, still 34.9% below baseline but representing meaningful improvement. Upload speeds recovered to 11.0 Mbps (down 29.7%), while latency moderated to 36 ms (up 8.3%). Notably, jitter remained stubbornly elevated at 9 ms (up 21.6%), suggesting that while raw throughput was improving, network stability had not yet fully normalized.

Starlink as a Resilience Layer

As mobile network performance declined, Speedtest data reveals a striking and sustained surge in Starlink usage across Portugal, providing one of the clearest real-world illustrations of satellite connectivity functioning as a resilience layer during prolonged terrestrial disruption.

In the weeks before Storm Kristin, Starlink activity in Portugal was relatively stable. From January 28 onward, however, user activity began climbing sharply. During the acute phase from January 28 through January 31, Starlink user activity averaged 49.4% above baseline, peaking at 61.3% above on January 31 as the scale of mobile network disruption became apparent. This initial surge likely reflects both existing Starlink subscribers increasing their usage in response to degraded mobile and fixed service and new users activating service for the first time.

The sustained disruption phase from February 1 through February 14 saw Starlink activity more than double, averaging 118.4% above baseline. The single highest day came on February 12, when user activity reached approximately 196% above pre-storm levels. This coincided with the period of deepest mobile network degradation (and, likely, fixed network unavailability either due to localized power loss or line faults), providing strong evidence of a network substitution dynamic where some users turned to satellite connectivity as their primary or sole means of internet access.

Perhaps most notably, Starlink user activity did not recede even as mobile networks began their partial recovery. During the February 15 through February 23 period, Starlink activity averaged 151.0% above baseline, substantially higher than even the acute storm phase. This pattern suggests that for many users, the storm experience catalyzed a longer-term shift in connectivity behavior, with satellite maintained as either a primary or backup connection even after terrestrial alternatives began stabilizing.

Portuguese authorities also actively deployed Starlink as an emergency communications tool. Starlink equipment was distributed to remote areas where traditional telecommunications had been knocked offline, helping to bridge the connectivity gap in the hardest-hit communities. This mirrors the pattern observed during the April 2025 Iberian blackout, when Starlink remained operational across the peninsula by routing through ground stations in Italy as Spanish facilities went dark.

It is worth noting that Starlink speeds did moderate as user load increased. Average download speeds during the sustained phase fell to 163.5 Mbps, a 21.7% decline from the pre-storm Starlink baseline of 208.8 Mbps. However, even at their most congested, Starlink speeds remained materially higher than the degraded mobile network’s performance during the same period, delivering nearly three times the median download speed that mobile users were experiencing.

The Regulatory Gap and the Road to Resilience

The storm sequence reinforces a core resilience lesson: in prolonged extreme-weather events, telecom continuity is heavily shaped by power autonomy at sites, restoration logistics, and transport network redundancy, not only by RAN capacity. In practice, the biggest outages often reflect cross-sector interdependence between electricity, fibre transport and mobile access infrastructure.

Indeed, domestically, the storms have reignited debate around the resilience of Portugal’s telecom infrastructure, particularly the adequacy of backup power provisions at mobile sites. The finding that 40% of failures stemmed from power loss at mobile sites, rather than direct storm damage, points to a structural vulnerability that is within regulatory reach to address.

Portugal currently lacks binding requirements for specific minimum backup power autonomy levels at mobile sites. This stands in contrast to Nordic markets such as Norway and Finland, where regulators require between two and six hours of backup power at critical sites, alongside routine stress testing and contingency planning obligations.

In Norway, Nkom’s forsterket ekom programme is a state-backed resilience scheme that hardens selected sites in priority municipalities. Designated mobile sites must have at least 72 hours of backup power, the main transmission path must also have 72 hours, and a separate reserve transmission path is required. Switzerland has also recently codified a phased minimum backup approach. In January 2026, the Federal Council adopted an FDV revision requiring mobile operators to install emergency power at key sites and antennas so mobile service can be maintained for at least four hours from 2031 (with emergency calls covered first, and other services phased in later).

Within the EU, meanwhile, Brussels’ Electronic Communications Code (EECC) permits member states to mandate such provisions but does not require them. The Commission’s DNA proposal, adopted on 21 January 2026, is framed around investment and simplification, but it also directly elevates resilience by introducing an EU-level Preparedness Plan to address rising risks from natural disasters and foreign interference, and by embedding security and resilience criteria into the pan-EU satellite mechanism.

The resilience policy implications arising from this are important. The DNA gives the EU a stronger coordination spine for preparedness, but it does not remove the need for national regulators to set concrete, site-level resilience expectations (including backup-power minimums) that reflect local grid conditions and risk exposure.

On the operator side, Vodafone’s Enhanced Power initiative, launched in November 2025 with Portugal as a first deployment region, targets 10,000-plus mobile infrastructure sites across Europe with backup power provisions ranging from four hours at critical access sites to 72 hours at core mobile data centers. The initiative incorporates AI-based systems to predict and conserve backup power duration. Separately, Portugal’s government has announced a €400 (US$ 471) million investment package for grid resilience, including a 750 MW battery storage expansion.

The experience of January and February 2026 reinforces what the Storm Éowyn analysis across the UK and Ireland also demonstrated: that the resilience of mobile networks in extreme weather is fundamentally a function of power autonomy at mobile sites. Where terrestrial infrastructure falls short, satellite connectivity is increasingly proving its value, not as a replacement for mobile networks, but as a complementary layer of redundancy that can sustain connectivity when ground-based systems falter.

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.

| February 18, 2026

5G in the Balkans: Serbia’s Mobile Renaissance as New Spectrum Fuels Competitiveness

Serbian/Srpski

EXPO 2027 is acting as a forcing function for policy acceleration and infrastructure investment as Belgrade prepares to put Serbia and the Balkans on the world stage

Serbia has moved rapidly from one of the last 5G “white spots” on the European map to a strong global performer in mobile. Following years of regulatory delays, the country’s operators activated commercial 5G services simultaneously on December 3, 2025, marking the start of a significant turnaround in network performance. Within weeks of launch, Serbia climbed 44 places to 14th globally in the Speedtest Global Index™ for December 2025, posting median download speeds of 173.04 Mbps and outperforming many Western European markets with more mature 5G deployments.

While Serbia enters the 5G cycle years after many of its Balkan peers, the country now finds itself well positioned to close the gap with regional competitors. Serbian operators have developed a mature 4G footprint on which they are now overlaying 5G equipment that is a generation or two newer than that deployed by earlier movers elsewhere in Europe, and which they had already stockpiled in anticipation of the auction. Belgrade’s hosting of EXPO 2027 is reinforcing this momentum, acting as a catalyst for infrastructure investment and compelling operators to achieve in two years what regional peers accomplished in five.

Key Takeaways

  • Serbia’s 5G launch is delivering immediate and substantial performance gains. While delayed spectrum assignment had left Serbia second-to-last in median mobile download speeds among Balkan peers in Q3 2025, the December 2025 commercial launch triggered a sharp reversal. All three operators recorded triple-digit percentage increases in median download speeds during Q4 2025, with national speeds rising from ~62 Mbps to over 150 Mbps, propelling Serbia to 14th globally in the December issue of the Speedtest Global Index™.
  • Serbia’s rapid ascent validates the potential for late movers to outperform in 5G. Serbia now sits alongside Bulgaria (5th) and North Macedonia (17th) as evidence that Balkan markets can achieve strong mobile outcomes when sound spectrum policy is paired with aggressive operator execution. The 130 MHz of 3.5 GHz spectrum allocated to each operator provides the capacity headroom to sustain these gains as network load increases.
  • A1 leads on speed while Yettel shows the strongest improvement trajectory. A1 delivered the highest median mobile download speeds (208.29 Mbps) in Q4 2025, while Yettel recorded the largest relative improvement post-5G launch (+160% quarter-on-quarter to 138.45 Mbps). Telekom Srbija (mts) reached 109.02 Mbps, more than doubling its Q3 performance. All three operators achieved near-parity on latency at ~32 ms.
  • Telekom faces competitive pressure to accelerate its 5G execution. Despite leading in subscriber share, Telekom trails A1 and Yettel in download speeds, 4G Availability (90% versus 94-95% for peers), and 5G Availability (19% versus 24% for A1). Its ongoing vendor transition adds execution complexity during a critical deployment window.
  • Regional mobile disparities persist but coverage obligations aim to close the gap. The 5G rollout obligations imposed by Serbian regulator RATEL require operators to cover 15 explicitly named underserved municipalities, major transport corridors, and progressively smaller population centers over a five-year period, creating a policy-driven mechanism to extend performance gains beyond Belgrade and Novi Sad.

A long road to spectrum. How Serbia finally got to a 5G auction

From policy delays to late-mover advantage

Serbia adopted policy documents referencing 5G as early as 2018, but the original expectation of assigning “pioneer band” spectrum around 2020 to 2021 slipped repeatedly. COVID-19, macroeconomic uncertainty, and domestic political debate over market structure and the role of state-owned Telekom Srbija all contributed to delays, making Serbia the last country in the Balkans apart from Bosnia and Herzegovina to launch a 5G auction process.

Independent assessments by the OECD and EU enlargement services explicitly flagged delays in 5G spectrum assignment and incomplete supporting regulation, such as a dedicated broadband infrastructure law, as key impediments to Serbia’s digital competitiveness.

Serbia Behind Peers Until Late 2025
Speedtest Intelligence® | Q4 2024 – Q4 2025

Analysis of Speedtest Intelligence® data reveals these delays translated into an erosion of the country’s regional ranking in mobile performance through Q3 2025. Between 2020 and Q3 2025, Serbia’s ranking dropped from fifth to tenth among Balkan markets in median mobile download speed. This observation is consistent with our earlier research, which found that European countries that moved earlier on 5G spectrum assignment had achieved better network outcomes for end-users.

However, Serbia’s late entry also carries advantages that are now becoming apparent. Operators have been able to deploy more mature 5G equipment (atop a densified 4G site grid with deep carrier aggregation), benefiting from cost reductions and performance improvements accumulated over several generations of 5G hardware, while simultaneously activating large, clean spectrum blocks rather than the fragmented allocations that constrained some earlier movers.

The early Q4 2025 performance data suggests Serbia may be capturing a meaningful late-mover dividend. The jump to 14th globally, ahead of many Western European markets with multi-year head starts, indicates that the quality of spectrum assignments and equipment maturity can outweigh time-to-market in determining 5G outcomes.

Long license duration sets Serbia’s 5G auction apart

The legislative blockage holding up Serbia’s 5G auction eased with a new Law on Electronic Communications adopted in 2023, which aligned more closely with the EU’s European Electronic Communications Code (EECC) and mandated the Serbian regulator RATEL to prepare for a multi-band auction. A subsequent national electronic communications strategy and rulebook set out the conditions for issuing licenses across 700 MHz, sub-3 GHz bands and the 3.4-3.8 GHz range, with a stated goal of maximizing both revenue and coverage outcomes.

RATEL’s auction design established strict participation criteria that effectively limited bidders to incumbent operators, a notably different approach from some regulators elsewhere in Europe that used 5G auctions to stimulate competition by making provisions for new entrants.

The auction ran from November 3 through November 7, 2025, attracted all three incumbents and raised just over €300 million (US$349.14 million), with each operator bidding slightly above the €100 million reserve for its package. Licenses are to be paid in two installments, with the second due by mid-2026, and are valid until March 2047, giving operators a 20-plus-year runway that is notably more generous than the 15-year norms observed in much of the EU (which is now subject to change based on the proposals for indefinite licences detailed in the European Commission’s Digital Networks Act).

On closer inspection, these auction results suggest Serbia traded a longer license tenure for robust returns to the state, while leaving operators with a comparatively light annual spectrum burden and a cost position that looks broadly competitive once adjusted for band mix and license duration.

The approximate per-capita yield from Serbia’s auction (around €44 per head) was materially above earlier Western Balkan awards on a headline basis. Croatia’s 2021 700/3.6/26 GHz sale yielded €12 per head, Bulgaria’s 3.6 GHz-only auction yielded €1 per head for 20-year licenses, and Albania’s 3.5 GHz award came in at roughly €2 per head. However, those comparisons typically covered much less spectrum, shorter terms, or both. Normalizing for Serbia’s extended tenure and broad multi-band packages, the annualized $/MHz/pop figure sits in the lower half of the European 5G range, well below historic high-price outliers such as Italy, Germany, or the early US C-band auction.

Balanced 5G spectrum assignments position Serbia for sustained performance gains

Beyond a favorable license duration, another key strength of the Serbian auction was that it lent each operator a broadly balanced low- and mid-band portfolio for 4G and 5G at roughly €100 million each. All three secured symmetrical holdings in the critical 700 MHz and 3.5 GHz pioneer bands, alongside individual top-ups in 2600 MHz and other legacy bands like 900, 1800 and 2100 MHz. This symmetry ensures no player is structurally spectrum-constrained, meaning competitive advantage will hinge less on spectrum depth and more on how quickly each operator executes on RAN densification and core network rollout.

All three can build a classic European 5G stack, eventually using 700 MHz as the wide-area SA coverage layer, refarmed 900/1800/2100 MHz as the 4G/5G anchor and VoLTE bedrock, 2600 MHz for dense capacity, and a wide C-band carrier as the primary 5G capacity layer, fully aligned with EU pioneer band thinking.

The most notable award was the 130 MHz of prime 3.5 GHz spectrum issued to each operator, a highly competitive mid-band allocation by European standards, where many markets targeted only 80 to 100 MHz per operator in this band. This positions Serbian operators well to enhance network capacity in dense urban environments, and the early performance data from Q4 2025 confirms that this wide spectrum allocation is already translating into tangible speed advantages for end-users.

Targeted coverage obligations for underserved areas seek to stimulate better 5G outcomes

A key feature of the auction design was the inclusion of a targeted coverage obligation for 15 explicitly named underserved municipalities, intended to prevent a purely city-centric 5G rollout concentrated in Belgrade and Novi Sad. Each operator must materially improve coverage, including future 5G, in five rural or structurally weak areas. Telekom is responsible for Krupanj, Osečina, Medveđa, Trgovište and Bosilegrad. Yettel is responsible for Ljubovija, Golubac, Babušnica, Gadžin Han and Svrljig. A1 is responsible for Kosjerić, Crna Trava, Kučevo, Majdanpek and Boljevac.

Many of these areas are officially classified as underdeveloped, with aging populations, high emigration and weak infrastructure. Crna Trava, Serbia’s smallest and one of its poorest municipalities, is the most emblematic example. In practice, this policy intervention is intended to compel operators to use new low-band spectrum to extend basic 4G/5G availability to places that would otherwise be unlikely to receive commercial coverage in the near future.

At a higher level, these municipal carve-outs sit atop a demanding five-year national rollout regime set by RATEL. License conditions require each operator to deploy at least 200 active 5G sites in year one, with 120 in Belgrade, 20 in Novi Sad, 15 in Niš, 10 in Kragujevac, 5 in Subotica, plus all science and tech parks. This rises to 1,000 in year two across 22 designated cities, major tourist centers, all international airports, and 24-hour road border crossings.

Subsequent targets rise to 2,000, 3,300, and 4,500 sites in years three through five, with mandatory coverage of all settlements above 10,000, 5,000, and 3,000 inhabitants respectively. By end-2026, operators must also achieve near-continuous coverage along IA/IB-class state roads and pan-European rail corridors X and XI, plus the EXPO 2027 site and its access routes. Each operator must meet these milestones on its own independent RAN, as national roaming does not count, thereby substantially raising minimum capex requirements compared with more permissive EU regimes.

5G rollout arrives amid challenging operating conditions

Similar to the trend observed elsewhere in the Balkans, the financial backdrop underpinning Serbia’s mobile market is one of steady, mid-single digit revenue growth up to the end of this decade. Mobile average revenue per user (ARPU) in Serbia sits at the lower end of the European range but ranks fourth-highest in the Balkans, averaging around €9 in late 2025. This figure is down materially in real terms from two to three years prior.

This sluggish ARPU profile therefore constrains how much 5G investment can be recouped from topline expansion alone, with all operators facing several capex and opex pressures across license fees and spectrum amortization, energy costs and fiber and EXPO-related investments.

The €300 million (US $349.11 million) collected in the recent auction is material relative to the size of the Serbian market. Even though fees are spread over time, they still absorb balance sheet capacity. Serbian operators, like their European peers, have faced electricity price volatility since 2022, with energy accounting for a non-trivial share of opex. This creates strong incentives for deploying energy-efficient radios and using features such as 5G sleep modes as the buildout advances. The government has emphasized both rural broadband and preparations for EXPO 2027 in Belgrade as strategic priorities, requiring upgrades in backhaul, core and metro transport infrastructure.

Within this challenging environment, earlier market moves have helped to free up some capital for 5G. Telekom completed a tower sale to an Actis-led consortium, while A1 did similar with its EuroTeleSites spin-off. Meanwhile, PPF Telecom Group, in which Etisalat (e&) acquired a majority stake across the wider Central and Eastern European portfolio, decided to structurally separate Yettel’s network into CETIN Serbia. This entity now owns and operates the passive and active infrastructure and provides wholesale mobile network services to Yettel.

Strategic focus of Serbian operators is tilting from volume to value

Serbia’s three operators have diverged into distinct strategic postures. State-backed Telekom remains the volume leader with approximately 42-44% market share and generates the most absolute revenue (€1.3-1.4 billion across the group in 2024), but carries the heaviest leverage burden, trading free cash flow for aggressive content spending on Premier League and NBA rights to defend ARPU against churn. Fixed-mobile bundling further supports its subscriber stickiness, particularly among households attracted to premium sports content.

Incremental value creation in the market is, however, shifting toward the leaner challengers. Yettel has successfully pivoted from a mobile-only cash cow, boasting the highest ARPU (around €11.60 in 2024), to a converged challenger. Backed by the PPF Group’s efficiency focus, it maintains the market’s healthiest EBITDA margins (19% expansion in 2024 to €171 million or US $198.99 million) by aggressively cross-selling its ‘Hipernet’ fixed services to a loyal mobile base of roughly 30% market share, effectively insulating itself from price wars.

A1, still the smallest operator by market share at around 27%, remains the primary competitive challenger and is closing the profitability gap as A1 Group transforms its Serbian business into a convergent player. It has shed its discounter identity and positioned itself as a distinct value innovator with a lean model capable of deploying capital quickly.

The operator posted EBITDA expansion of 15% in 2024 to €146 million or US $169.90 million (margin of approximately 37%), driven by inflation-linked price rises and upselling to higher-value plans, even as prepaid and postpaid bases contracted. The Conexio Metro fiber acquisition (42,000 ready homes) and the launch of fixed services in 2025 is intended to further lift A1’s ARPU and reduce churn.

A1 and Yettel entered the 5G cycle with a stronger 4G network foundation

Serbia’s late 5G RAN cycle forced operators to carry surging data demand almost entirely on their 4G networks through most of 2025 pre-auction. Each operator turned to leveraging 800, 1800 and 2100 MHz refarming, three- to four-carrier aggregation and targeted grid densification along urban areas and transport corridors to maximize spatial spectrum reuse.

While this 4G-centric playbook pushed the existing grid far, the economics were deteriorating for operators. Each incremental site or radio upgrade was capex- and energy-heavy in a price-sensitive market, and spectral-efficiency gains from further LTE-Advanced tweaks were marginal compared with lighting up large, clean 5G carriers in the 3.5 GHz band.

The performance trajectory through 2025 highlights the different starting positions from which operators entered the 5G cycle. A1 had maintained a substantial speed lead in Serbia even on 4G, with median mobile download speeds on its network of 81.03 Mbps in Q3 2025, roughly 40% faster than Yettel (53.15 Mbps) and Telekom (57.95 Mbps).

Early 5G performance exceeds expectations as operators light up new spectrum

The simultaneous commercial launch of 5G by all three Serbian operators on December 3, 2025 triggered an immediate and significant transformation in network performance. Analysis of Speedtest Intelligence® data for Q4 2025 reveals speed increases far exceeding what would typically be observed from incremental 4G improvements, confirming that the newly activated 3.5 GHz and 700 MHz spectrum is already delivering substantial benefits to end-users.

A1 delivered median mobile download speeds of 208.29 Mbps in Q4 2025, an increase of 157% from 81.03 Mbps in Q3 2025. Yettel recorded the largest relative improvement, with median speeds rising 160% from 53.15 Mbps to 138.45 Mbps. Telekom, while posting the lowest absolute speeds among the three operators, still achieved an 88% increase to 109.02 Mbps from 57.95 Mbps.

A1 Extends Speed Lead as All MNOs Record 5G Gains
Speedtest Intelligence® | Q4 2024 – Q4 2025

Upload speeds similarly improved across all operators, with Yettel leading at 27.02 Mbps, followed by Telekom (24.02 Mbps) and A1 (23.41 Mbps). Multi-server latency converged to ~32 ms across all three networks, down from divergent Q3 figures.

Upload Speeds Improve Post-5G Launch With Yettel Taking the Lead
Speedtest Intelligence® | Q4 2024 – Q4 2025

The clearest evidence of Serbia’s ongoing mobile transformation is its rise in the Speedtest Global Index. In December 2025, Serbia climbed 44 positions to rank 14th globally with median download speeds of 173.04 Mbps, placing it ahead of established European 5G markets including Georgia (15th), Netherlands (16th), Norway (20th), France (26th), and Finland (27th). While some reversion is highly likely as network load increases through 2026, the wide 130 MHz spectrum allocation per operator provides uniquely substantial headroom to absorb traffic growth while maintaining strong performance.

Analysis of 5G Availability data confirms that operators activated commercial service rapidly following the December 3 launch. A1 led with 5G Availability of 24% in Q4 2025, indicating that users on 5G-capable devices on its network spent approximately one-quarter of their time connected to 5G. Yettel and Telekom recorded similar figures of 19%. These availability figures are expected to rise steadily as operators expand their 5G footprints beyond initial launch areas in Belgrade and major cities.

Analysis of 5G Availability data confirms that operators activated commercial service rapidly following the December 3 launch. A1 led with 5G Availability of 24% in Q4 2025, indicating that users on 5G-capable devices on its network spent approximately one-quarter of their time connected to 5G. Yettel and Telekom recorded similar figures of 19%. These availability figures are expected to rise steadily as operators expand their 5G footprints beyond initial launch areas in Belgrade and major cities.

Divergence in 3G sunset progress highlights differences in spectrum strategy and 4G maturity

Investments in site grid densification and 4G footprint expansion have contributed to improved network consistency and coverage outcomes across all operators. 4G Availability in Q4 2025 reached 95% for A1, 94% for Yettel, and 90% for Telekom, reflecting wide coverage in rural areas and extensive time spent on 4G (and now 5G) networks indoors in urban areas as low-band expansion and site densification aid deeper building penetration in cities like Belgrade and Novi Sad.

A1's Broader 4G Footprint Provides Stronger Foundation for 5G Overlay
Speedtest Intelligence® | Q4 2024 – Q4 2025

The wider breadth of A1’s 4G network and extensive use of VoLTE across its footprint enabled it to move more aggressively than peers on sunsetting 3G services. After announcing a phased 3G switch-off by end-June 2025, A1 completed nationwide 3G decommissioning in April 2025, freeing its 2100 MHz 3G layer for 4G/5G while explicitly keeping 2G in the 900 MHz band alive as a thin legacy voice/SMS and low-end IoT anchor and pushing all mainstream voice onto VoLTE.

A1’s 3G sunset has effectively converted its 2100 MHz footprint into a pure 4G capacity layer, while its relatively small 4.2 MHz slice of 900 MHz (versus 9.6 MHz for Telekom and Yettel) makes it structurally incentivized to accelerate VoLTE migration and shrink 2G usage over time.

Yettel, by contrast, has no formal 2G/3G sunset dates but has already silently refarmed much of its original 2100 MHz 3G layer to 4G while keeping a low-band dual 2G/3G umbrella across mostly 900 MHz until VoLTE use and the 5G rollout mature.

Telekom remains the most conservative on network sunset timing, reflecting a larger base of rural and legacy users and a desire to avoid a voice cliff while VoLTE usage still scales atop a comparatively smaller 4G coverage base. The operator continues to run both 2G (900/1800 MHz) and 3G (primarily 2100 MHz) alongside 4G/5G.

Vendor strategies diverge as operators navigate equipment transitions

Beyond the benefit of having a wide 4G footprint, the speed at which A1 has been able to move with its 3G sunset is likely shaped in some way by its uniquely clean vendor strategy. A1 Group has already locked Serbia into a single-RAN strategy with Nokia supplying both radio and packet core, under a five-year 5G plan to cover major population centers and transport corridors, using EuroTeleSites’ spun-off tower grid to keep the RAN relatively asset-light.

Telekom, by contrast, is in mid-pivot from a Huawei-heavy legacy 2G-4G footprint to a dual-Nordic 5G layer, financed through a distinctive multilateral structure including the European Bank for Reconstruction and Development, Sweden’s Export Credit Corporation and the US Exim Bank. Notably, the US Exim deal of approximately €42.97 million or US $50 million represents the first agreement of its kind between a European telecom operator and the US Export-Import Bank. Ericsson and Nokia are contracted to supply 5G base stations for the network rollout.

Yettel, through CETIN Serbia, is also undergoing active vendor diversification as it shifts from Huawei, used historically for its RAN, to Ericsson for its 5G buildout. This transition is being coordinated alongside CETIN’s broader “5G Balkans” project, which aims to strengthen optical connectivity across the region in partnership with CETIN Bulgaria.

5G boost comes just in time for EXPO 2027

The simultaneous 5G launch in December 2025 by all three operators, backed by generous 130 MHz mid-band allocations and modern equipment that had been stockpiled during the regulatory delays, has propelled Serbia from near the bottom of the Balkan mobile rankings to a highly competitive position globally, validating the potential for late movers to close the gap with competitors when spectrum policy and operator execution align.

With EXPO 2027 on the horizon and the government committed to using the event as a showcase for the country’s digital capabilities, Serbia’s mobile operators face both competitive pressure and a significant opportunity to demonstrate what investments in 5G infrastructure can deliver for consumers and businesses alike. The early data suggests they are rising to the challenge.

Serbian/Srpski


5G na Balkanu. Mobilna renesansa Srbije dok novi spektar podstiče konkurentnost

EXPO 2027 deluje kao ključni pokretač za ubrzanje politike i infrastrukturnih ulaganja, dok se Beograd priprema da plasira Srbiju i Balkan na svetsku pozornicu.

Srbija je hitro prešla put od jedne od poslednjih 5G “belih tačaka” na mapi Evrope do snažnog globalnog igrača u mobilnoj telefoniji. Nakon godina regulatornih odlaganja, operatori u zemlji su 3. decembra 2025. godine istovremeno aktivirali komercijalne 5G servise, označivši početak značajnog preokreta u performansama mreže. U roku od nekoliko nedelja nakon aktiviranja, Srbija je napredovala za 44 mesta i zauzela 14. poziciju na globalnom nivou u Speedtest Global Index™ za decembar 2025. godine, zabeleživši medijanu brzine preuzimanja od 173,04 Mbps i nadmašivši mnoga zapadnoevropska tržišta sa zrelijim 5G mrežama.

Iako Srbija ulazi u 5G ciklus godinama nakon mnogih svojih balkanskih suseda, zemlja se sada nalazi u odličnoj poziciji da smanji razliku u odnosu na regionalne konkurente. Srpski operatori su razvili zrelu 4G osnovu na koju sada nadograđuju 5G opremu koja je generaciju ili dve novija od one koju su primenili rani usvajaoci (early movers) drugde u Evropi, a koju su već bili pripremili u iščekivanju aukcije. EXPO 2027 koji će se održati u Beogradu dodatno pojačava ovaj zamah, delujući kao katalizator za infrastrukturna ulaganja i primoravajući operatore da za dve godine postignu ono za šta je regionalnim konkurentima bilo potrebno pet.

Ključne poruke

  • Lansiranje 5G mreže u Srbiji donosi trenutna i značajna poboljšanja performansi. Iako je odložena dodela spektra ostavila Srbiju na pretposlednjem mestu po medijani brzine mobilnog preuzimanja među balkanskim zemljama u trećem kvartalu 2025., komercijalno lansiranje u decembru 2025. izazvalo je oštar preokret. Sva tri operatora zabeležila su trocifreni procentualni rast medijane brzine preuzimanja tokom četvrtog kvartala 2025., pri čemu su nacionalne brzine porasle sa ~62 Mbps na preko 150 Mbps, što je Srbiju lansiralo na 14. mesto globalno u decembarskom izdanju Speedtest Global Index™.
  • Brzi uspon Srbije potvrđuje potencijal kasnih ulazaka na 5G tržište (late movers) da nadmaše očekivanja. Srbija sada stoji rame uz rame sa Bugarskom (5. mesto) i Severnom Makedonijom (17. mesto) kao dokaz da balkanska tržišta mogu postići izvrsne rezultate u mobilnoj telefoniji kada se zdrava politika spektra upari sa agresivnom realizacijom od strane operatora. Dodeljenih 130 MHz spektra u opsegu od 3,5 GHz po operatoru pruža dovoljno kapaciteta za održavanje ovih rezultata kako opterećenje mreže bude raslo.
  • A1 prednjači u brzini, dok Yettel pokazuje najsnažniji trend poboljšanja. A1 je isporučio najveću medijanu brzine mobilnog preuzimanja (208,29 Mbps) u četvrtom kvartalu 2025., dok je Yettel zabeležio najveće relativno poboljšanje nakon lansiranja 5G mreže (+160% u odnosu na prethodni kvartal, na 138,45 Mbps). Telekom Srbija (mts) je dostigao 109,02 Mbps, više nego udvostručivši svoj rezultat iz trećeg kvartala. Sva tri operatora su postigla gotovo izjednačene rezultate u kašnjenju (latenciji) od oko 32 ms.
  • Telekom se suočava sa konkurentskim pritiskom da ubrza svoju 5G implementaciju. Uprkos tome što vodi u udelu pretplatnika, Telekom zaostaje za A1 i Yettel-om u brzinama preuzimanja, dostupnosti 4G mreže (90% naspram 94-95% kod konkurenata) i dostupnosti 5G mreže (19% naspram 24% za A1). Njegova trenutna tranzicija dobavljača opreme dodaje složenost u realizaciji tokom ključnog perioda izgradnje mreže.
  • Regionalne razlike u mobilnim uslugama i dalje postoje, ali obaveze pokrivanja imaju za cilj da smanje jaz. Obaveze uvođenja 5G mreže koje je nametnuo srpski regulator RATEL zahtevaju od operatora da pokriju 15 eksplicitno navedenih nedovoljno razvijenih opština, glavne transportne koridore i postepeno manja naselja tokom petogodišnjeg perioda, stvarajući mehanizam vođen politikom za proširenje poboljšanja performansi i van Beograda i Novog Sada.

Dug put do spektra: Kako je Srbija konačno stigla do 5G aukcije

Od kašnjenja politike do prednosti kasnog ulaska

Srbija je usvojila strateška dokumenta koja pominju 5G još 2018. godine, ali je prvobitno očekivanje o dodeli spektra u “pionirskim opsezima” oko 2020. ili 2021. godine više puta odlagano. COVID-19, makroekonomska neizvesnost i unutrašnja politička debata o strukturi tržišta i ulozi državnog Telekoma Srbija doprineli su kašnjenju, čineći Srbiju poslednjom zemljom na Balkanu, osim Bosne i Hercegovine, koja je pokrenula proces 5G aukcije.

Nezavisne procene OECD-a i službi za proširenje EU eksplicitno su označile kašnjenja u dodeli 5G spektra i nepotpunu prateću regulativu, kao što je namenski zakon o širokopojasnoj infrastrukturi, kao ključne prepreke za digitalnu konkurentnost Srbije.

Analiza podataka Speedtest Intelligence otkriva da su se ova kašnjenja prevela u eroziju regionalnog rangiranja zemlje u mobilnim performansama sve do trećeg kvartala 2025. Između 2020. i trećeg kvartala 2025., rang Srbije je pao sa petog na deseto mesto među balkanskim tržištima po medijani brzine mobilnog preuzimanja. Ovo zapažanje je u skladu sa našim ranijim istraživanjima, koja su utvrdila da su evropske zemlje koje su ranije dodelile 5G spektar postigle bolje mrežne ishode za krajnje korisnike.

Srbija skočila sa dna na treće mesto među balkanskim zemljama nakon pokretanja 5G mreže
Speedtest Intelligence® | Q4 2024 – Q4 2025

Međutim, kasni ulazak Srbije nosi i prednosti koje sada postaju očigledne. Operatori su bili u mogućnosti da primene zreliju 5G opremu (povrh već zgusnute 4G mreže sa dubokom agregacijom nosilaca), koristeći smanjenje troškova i poboljšanja performansi akumulirana kroz nekoliko generacija 5G hardvera, istovremeno aktivirajući velike, čiste blokove spektra umesto fragmentisanih alokacija koje su ograničavale neke ranije usvajaoce.

Podaci o performansama s početka četvrtog kvartala 2025. sugerišu da Srbija možda ostvaruje značajnu “dividendu kasnog ulaska”. Skok na 14. mesto globalno, ispred mnogih zapadnoevropskih tržišta sa višegodišnjom prednošću, ukazuje na to da kvalitet dodele spektra i zrelost opreme mogu nadvladati vreme izlaska na tržište (time-to-market) u određivanju 5G ishoda.

Dugo trajanje licenci izdvaja srpsku 5G aukciju

Zakonodavna blokada koja je kočila srpsku 5G aukciju popustila je usvajanjem novog Zakona o elektronskim komunikacijama 2023. godine, koji se bliže uskladio sa Evropskim zakonikom o elektronskim komunikacijama (EECC) EU i naložio srpskom regulatoru RATEL-u da se pripremi za aukciju više opsega (multi-band auction). Naknadna nacionalna strategija elektronskih komunikacija i pravilnik postavili su uslove za izdavanje licenci u opsezima od 700 MHz, ispod 3 GHz i opsegu 3,4–3,8 GHz, sa proklamovanim ciljem maksimiziranja i prihoda i pokrivenosti.

RATEL-ov dizajn aukcije uspostavio je stroge kriterijume za učešće koji su efektivno ograničili ponuđače na postojeće operatore, što je znatno drugačiji pristup od nekih regulatora drugde u Evropi koji su koristili 5G aukcije da podstaknu konkurenciju stvaranjem uslova za nove učesnike (new entrants).

Aukcija je trajala od 3. do 7. novembra 2025., privukla je sva tri postojeća operatora i prikupila nešto više od 300 miliona evra (349,14 miliona dolara), pri čemu je svaki operator licitirao nešto iznad početne cene od 100 miliona evra za svoj paket. Licence se plaćaju u dve rate, sa drugom koja dospeva sredinom 2026. godine, i važe do marta 2047., dajući operatorima period od preko 20 godina, što je znatno velikodušnije od normi od 15 godina viđenih u većem delu EU (što je sada podložno promenama na osnovu predloga za trajne licence detaljno opisanih u Zakonu o digitalnim mrežama Evropske komisije).

Detaljnijom analizom, ovi rezultati aukcije sugerišu da je Srbija trampila duže trajanje licenci za robusne prihode državi, ostavljajući operatore sa relativno lakim godišnjim opterećenjem za spektar i troškovnom pozicijom koja izgleda široko konkurentna kada se prilagodi miksu opsega i trajanju licence.

Približan prinos po glavi stanovnika od srpske aukcije (oko 44 evra po stanovniku) bio je materijalno iznad ranijih dodela na Zapadnom Balkanu na nominalnoj osnovi. Hrvatska prodaja 700/3,6/26 GHz iz 2021. donela je 12 evra po glavi stanovnika, bugarska aukcija samo za 3,6 GHz donela je 1 evro po glavi stanovnika za 20-godišnje licence, a albanska dodela 3,5 GHz iznosila je otprilike 2 evra po glavi stanovnika. Međutim, ta poređenja su obično pokrivala mnogo manje spektra, kraće rokove ili oboje. Normalizovano za produženi rok trajanja u Srbiji i široke pakete više opsega, godišnja cifra $/MHz/pop nalazi se u donjoj polovini evropskog 5G raspona, znatno ispod istorijskih skupih izuzetaka kao što su Italija, Nemačka ili rana američka aukcija C opsega.

Uravnotežene dodele 5G spektra pozicioniraju Srbiju za održiva poboljšanja performansi

Pored povoljnog trajanja licence, još jedna ključna snaga srpske aukcije bila je to što je svakom operatoru dodelila široko uravnotežen portfolio niskih i srednjih opsega za 4G i 5G po ceni od otprilike 100 miliona evra svakom. Sva tri operatora su obezbedila simetrične udele u kritičnim pionirskim opsezima od 700 MHz i 3,5 GHz, uz pojedinačne dopune u opsegu od 2600 MHz i drugim nasleđenim opsezima poput 900, 1800 i 2100 MHz. Ova simetrija osigurava da nijedan igrač nije strukturno ograničen spektrom, što znači da će konkurentska prednost manje zavisiti od dubine spektra, a više od toga koliko brzo svaki operator realizuje zgušnjavanje RAN mreže (radio pristupne mreže) i uvođenje jezgra mreže (core network).

Sva tri operatora mogu da izgrade klasičan evropski 5G “stack” (slojevitu strukturu), koristeći 700 MHz kao sloj za široku SA pokrivenost, refarmirane opsege 900/1800/2100 MHz kao 4G/5G sidro i osnovu za VoLTE, 2600 MHz za gust kapacitet i široki C opseg kao primarni sloj 5G kapaciteta, potpuno usklađeno sa razmišljanjem EU o pionirskim opsezima.

Najznačajnija dodela bila je 130 MHz primarnog spektra od 3,5 GHz izdatog svakom operatoru, što je visoko konkurentna alokacija srednjeg opsega po evropskim standardima, gde su mnoga tržišta ciljala samo 80 do 100 MHz po operatoru u ovom opsegu. Ovo pozicionira srpske operatore u dobru situaciju da poboljšaju mrežni kapacitet u gustim urbanim sredinama, a rani podaci o performansama iz četvrtog kvartala 2025. potvrđuju da se ova široka alokacija spektra već prevodi u konkretne prednosti u brzini za krajnje korisnike.

Ciljane obaveze pokrivanja za nedovoljno razvijena područja teže da podstaknu bolje 5G ishode

Ključna karakteristika dizajna aukcije bilo je uključivanje ciljane obaveze pokrivanja za 15 eksplicitno imenovanih nedovoljno razvijenih opština, sa namerom da se spreči čisto “gradocentrično” uvođenje 5G mreže koncentrisano u Beogradu i Novom Sadu. Svaki operator mora materijalno poboljšati pokrivenost, uključujući budući 5G, u pet ruralnih ili strukturno slabih područja. Telekom je odgovoran za Krupanj, Osečinu, Medveđu, Trgovište i Bosilegrad. Yettel je odgovoran za Ljuboviju, Golubac, Babušnicu, Gadžin Han i Svrljig. A1 je odgovoran za Kosjerić, Crnu Travu, Kučevo, Majdanpek i Boljevac.

Mnoge od ovih oblasti su zvanično klasifikovane kao nerazvijene, sa starim stanovništvom, visokom emigracijom i slabom infrastrukturom. Crna Trava, najmanja i jedna od najsiromašnijih opština u Srbiji, najslikovitiji je primer. U praksi, ova intervencija politike ima za cilj da primora operatore da koriste novi spektar niskog opsega kako bi proširili osnovnu 4G/5G dostupnost na mesta koja inače verovatno ne bi dobila komercijalnu pokrivenost u bliskoj budućnosti.

Na višem nivou, ova opštinska izdvajanja dolaze povrh zahtevnog petogodišnjeg nacionalnog režima izgradnje koji je postavio RATEL. Uslovi licence zahtevaju da svaki operator postavi najmanje 200 aktivnih 5G baznih stanica u prvoj godini, od toga 120 u Beogradu, 20 u Novom Sadu, 15 u Nišu, 10 u Kragujevcu, 5 u Subotici, plus sve naučno-tehnološke parkove. Ovaj broj raste na 1.000 u drugoj godini širom 22 određena grada, glavnih turističkih centara, svih međunarodnih aerodroma i 24-časovnih drumskih graničnih prelaza.

Naknadni ciljevi rastu na 2.000, 3.300 i 4.500 lokacija u trećoj do pete godini, sa obaveznom pokrivenošću svih naselja iznad 10.000, 5.000 i 3.000 stanovnika, respektivno. Do kraja 2026., operatori takođe moraju postići gotovo neprekidnu pokrivenost duž državnih puteva IA/IB klase i panevropskih železničkih koridora X i XI, kao i lokacije EXPO 2027 i njenih pristupnih puteva. Svaki operator mora ispuniti ove prekretnice na sopstvenoj nezavisnoj RAN mreži, jer se nacionalni roming ne računa, čime se znatno podižu minimalni zahtevi za kapitalnim ulaganjima (capex) u poređenju sa popustljivijim režimima u EU.

Uvođenje 5G stiže usred izazovnih uslova poslovanja

Slično trendu primećenom drugde na Balkanu, finansijska pozadina koja podupire srpsko mobilno tržište je stabilan, jednocifren rast prihoda (mid-single digit) do kraja ove decenije. Prosečan prihod po korisniku mobilne telefonije (ARPU) u Srbiji nalazi se na donjem kraju evropskog raspona, ali je četvrti po visini na Balkanu, prosečno oko 9 evra krajem 2025. Ova cifra je materijalno niža u realnim vrednostima u odnosu na dve do tri godine ranije.

Ovaj spori profil ARPU-a stoga ograničava koliko se investicija u 5G može povratiti samo kroz rast prihoda, pri čemu se svi operatori suočavaju sa nekoliko pritisaka na kapitalne (capex) i operativne troškove (opex) kroz naknade za licence, amortizaciju spektra, troškove energije, kao i ulaganja u optiku i projekte vezane za EXPO.

Iznos od 300 miliona evra (349,11 miliona dolara) prikupljen na nedavnoj aukciji je značajan u odnosu na veličinu srpskog tržišta. Iako su naknade raspoređene tokom vremena, one i dalje apsorbuju kapacitet bilansa stanja. Srpski operatori, kao i njihove evropske kolege, suočili su se sa volatilnošću cena električne energije od 2022. godine, pri čemu energija čini značajan deo operativnih troškova. Ovo stvara snažne podsticaje za primenu energetski efikasnih radio uređaja i korišćenje funkcija kao što su 5G “režimi spavanja” (sleep modes) kako izgradnja napreduje. Vlada je istakla i ruralni širokopojasni internet i pripreme za EXPO 2027 u Beogradu kao strateške prioritete, zahtevajući nadogradnje u pozadinskoj mreži (backhaul), jezgru i metro transportnoj infrastrukturi.

U ovom izazovnom okruženju, raniji potezi na tržištu pomogli su da se oslobodi deo kapitala za 5G. Telekom je završio prodaju tornjeva konzorcijumu koji predvodi Actis, dok je A1 učinio slično sa svojim izdvajanjem EuroTeleSites kompanije. U međuvremenu, PPF Telecom Group, u kojoj je Etisalat (e&) stekao većinski udeo u širem portfoliju centralne i istočne Evrope, odlučio je da strukturno odvoji mrežu Yettel-a u CETIN Srbija. Ovaj entitet sada poseduje i upravlja pasivnom i aktivnom infrastrukturom i pruža veleprodajne usluge mobilne mreže Yettel-u.

Strateški fokus srpskih operatora prelazi sa obima na vrednost

Tri srpska operatora su se razišla u različite strateške pozicije. Telekom Srbija, podržan od strane države, ostaje lider u obimu sa približno 42-44% tržišnog udela i generiše najveći apsolutni prihod (1,3-1,4 milijarde evra na nivou grupe u 2024.), ali nosi najteži teret zaduženosti, trgujući slobodnim novčanim tokovima za agresivno trošenje na sadržaj (Premier liga i NBA prava) kako bi odbranio ARPU od odliva korisnika (churn). Fiksno-mobilno povezivanje dodatno podržava vezanost njegovih pretplatnika, posebno među domaćinstvima privučenim premium sportskim sadržajem.

Inkrementalno stvaranje vrednosti na tržištu, međutim, pomera se ka efikasnijim izazivačima. Yettel je uspešno prešao put od mobilne “krave muzare” (cash cow), hvaleći se najvišim ARPU-om (oko 11,60 evra u 2024.), do konvergentnog izazivača. Podržan fokusom PPF Grupe na efikasnost, održava najzdravije EBITDA marže na tržištu (rast od 19% u 2024. na 171 milion evra ili 198,99 miliona dolara) agresivnom unakrsnom prodajom svojih ‘Hipernet’ fiksnih usluga lojalnoj bazi mobilnih korisnika od oko 30% tržišnog udela, efikasno se izolujući od ratova cena.

A1, i dalje najmanji operator po tržišnom udelu sa oko 27%, ostaje primarni konkurentski izazivač i smanjuje jaz u profitabilnosti dok A1 Grupa transformiše svoje poslovanje u Srbiji u konvergentnog igrača. Odbacio je identitet diskontera i pozicionirao se kao poseban inovator vrednosti sa “vitkim” modelom sposobnim za brzo raspoređivanje kapitala.

Operator je zabeležio rast EBITDA od 15% u 2024. na 146 miliona evra ili 169,90 miliona dolara (marža od približno 37%), vođen rastom cena usklađenim sa inflacijom i prodajom skupljih paketa (upselling), čak i dok su se baze pripejd i postpejd korisnika smanjivale. Akvizicija Conexio Metro optike (42.000 spremnih domova) i lansiranje fiksnih usluga 2025. godine imaju za cilj da dodatno podignu ARPU A1 i smanje odliv korisnika.

A1 i Yettel ušli su u 5G ciklus sa jačom 4G mrežnom osnovom

Kasni ciklus 5G RAN-a u Srbiji primorao je operatore da rastuću potražnju za podacima nose gotovo u potpunosti na svojim 4G mrežama tokom većeg dela 2025. godine pre aukcije. Svaki operator se okrenuo korišćenju refarminga frekvencija 800, 1800 i 2100 MHz, agregaciji tri do četiri nosioca i ciljanom zgušnjavanju mreže duž urbanih područja i transportnih koridora kako bi maksimizirao prostornu ponovnu upotrebu spektra.

Iako je ova strategija fokusirana na 4G gurnula postojeću mrežu do krajnjih granica, ekonomija je za operatore postajala sve gora. Svaka dodatna lokacija ili nadogradnja radija bila je teška za kapitalne i energetske troškove na tržištu osetljivom na cene, a dobici u spektralnoj efikasnosti od daljih LTE-Advanced podešavanja bili su marginalni u poređenju sa aktiviranjem velikih, čistih 5G nosilaca u opsegu 3,5 GHz.

Putanja performansi tokom 2025. naglašava različite startne pozicije sa kojih su operatori ušli u 5G ciklus. A1 je zadržao značajnu prednost u brzini u Srbiji čak i na 4G, sa medijanom brzine mobilnog preuzimanja na svojoj mreži od 81,03 Mbps u trećem kvartalu 2025., što je otprilike 40% brže od Yettel-a (53,15 Mbps) i Telekoma (57,95 Mbps).

Rane 5G performanse premašuju očekivanja dok operatori aktiviraju novi spektar

Istovremeno komercijalno lansiranje 5G od strane sva tri srpska operatora 3. decembra 2025. izazvalo je trenutnu i značajnu transformaciju u performansama mreže. Analiza podataka Speedtest Intelligence za četvrti kvartal 2025. otkriva povećanja brzine koja daleko prevazilaze ono što bi se obično primetilo od inkrementalnih 4G poboljšanja, potvrđujući da novoaktivirani spektar od 3,5 GHz i 700 MHz već donosi značajne benefite krajnjim korisnicima.

A1 je isporučio medijanu brzine mobilnog preuzimanja od 208,29 Mbps u četvrtom kvartalu 2025., što je povećanje od 157% sa 81,03 Mbps u trećem kvartalu. Yettel je zabeležio najveće relativno poboljšanje, sa brzinama koje su porasle za 160% sa 53,15 Mbps na 138,45 Mbps. Telekom, iako beleži najniže apsolutne brzine među tri operatora, ipak je postigao povećanje od 88% na 109,02 Mbps sa 57,95 Mbps.

A1 proširuje prednost u brzini dok sva tri operatora beleže oštar rast zahvaljujući 5G
Speedtest Intelligence® | Q4 2024 – Q4 2025

Brzine otpremanja (upload) su se slično poboljšale kod svih operatora, pri čemu Yettel vodi sa 27,02 Mbps, a slede ga Telekom (24,02 Mbps) i A1 (23,41 Mbps). Latencija na više servera (multi-server latency) konvergirala je na ~32 ms na sve tri mreže, što je pad u odnosu na divergentne brojke iz trećeg kvartala.

Brzine otpremanja rastu nakon pokretanja 5G mreže, Yettel preuzima vođstvo
Speedtest Intelligence® | Q4 2024 – Q4 2025

Najjasniji dokaz tekuće mobilne transformacije Srbije je njen uspon na Speedtest Global Index-u. U decembru 2025., Srbija se popela za 44 pozicije na 14. mesto globalno sa medijanom brzine preuzimanja od 173,04 Mbps, plasirajući se ispred utvrđenih evropskih 5G tržišta uključujući Gruziju (15.), Holandiju (16.), Norvešku (20.), Francusku (26.) i Finsku (27.). Iako je određeni povratak nazad vrlo verovatan kako se opterećenje mreže bude povećavalo tokom 2026. godine, široka alokacija spektra od 130 MHz po operatoru pruža jedinstveno značajan prostor za apsorpciju rasta saobraćaja uz održavanje snažnih performansi.

Analiza podataka o dostupnosti 5G potvrđuje da su operatori brzo aktivirali komercijalnu uslugu nakon lansiranja 3. decembra. A1 je prednjačio sa dostupnošću 5G od 24% u četvrtom kvartalu 2025., što ukazuje da su korisnici na 5G uređajima u njegovoj mreži proveli približno četvrtinu svog vremena povezani na 5G. Yettel i Telekom su zabeležili slične brojke od 19%. Očekuje se da će ove brojke o dostupnosti stalno rasti kako operatori budu širili svoje 5G otiske izvan početnih zona lansiranja u Beogradu i većim gradovima.

Razlike u progresu gašenja 3G mreže ističu razlike u strategiji spektra i zrelosti 4G mreže

Ulaganja u zgušnjavanje mreže baznih stanica i proširenje 4G pokrivenosti doprinela su poboljšanoj konzistentnosti mreže i ishodima pokrivenosti kod svih operatora. Dostupnost 4G u četvrtom kvartalu 2025. dostigla je 95% za A1, 94% za Yettel i 90% za Telekom, odražavajući široku pokrivenost u ruralnim područjima i značajno vreme provedeno na 4G (a sada i 5G) mrežama u zatvorenom prostoru u urbanim područjima, kako širenje niskog opsega i zgušnjavanje lokacija pomažu dublju penetraciju u zgrade u gradovima poput Beograda i Novog Sada.

Širi obim 4G mreže A1 i opsežna upotreba VoLTE-a širom njegove pokrivenosti omogućili su mu da deluje agresivnije od konkurenata po pitanju gašenja (sunset) 3G usluga. Nakon najave faznog gašenja 3G do kraja juna 2025., A1 je završio ukidanje 3G mreže na nacionalnom nivou u aprilu 2025., oslobađajući svoj 3G sloj od 2100 MHz za 4G/5G, dok je eksplicitno zadržao 2G u opsegu 900 MHz kao tanko sidro za govorne/SMS usluge i IoT uređaje niže klase, gurajući sav glavni govorni saobraćaj na VoLTE.

Šira 4G pokrivenost A1 pruža čvršću osnovu za 5G nadogradnju
Speedtest Intelligence® | Q4 2024 – Q4 2025

Gašenje 3G mreže od strane A1 efektivno je konvertovalo njegov otisak od 2100 MHz u čisti sloj 4G kapaciteta, dok ga njegov relativno mali deo od 4,2 MHz u opsegu 900 MHz (u poređenju sa 9,6 MHz za Telekom i Yettel) čini strukturno motivisanim da ubrza migraciju na VoLTE i vremenom smanji upotrebu 2G mreže.

Yettel, nasuprot tome, nema formalne datume gašenja 2G/3G mreže, ali je već “tiho” prenamenio veliki deo svog originalnog 3G sloja od 2100 MHz u 4G, zadržavajući “kišobran” niskog dualnog opsega 2G/3G uglavnom na 900 MHz dok upotreba VoLTE-a i uvođenje 5G mreže ne sazru.

Telekom ostaje najkonzervativniji po pitanju tajminga gašenja mreže, što odražava veću bazu ruralnih i starih korisnika i želju da se izbegne nagli prekid glasovnih usluga dok upotreba VoLTE-a još uvek raste povrh komparativno manje baze 4G pokrivenosti. Operator nastavlja da koristi i 2G (900/1800 MHz) i 3G (prvenstveno 2100 MHz) uporedo sa 4G/5G.

Strategije dobavljača se razlikuju dok operatori upravljaju tranzicijom opreme

Pored prednosti posedovanja širokog 4G otiska, brzina kojom je A1 mogao da sprovede gašenje 3G verovatno je na neki način oblikovana njegovom jedinstveno čistom strategijom dobavljača. A1 Grupa je već zaključala Srbiju u single-RAN strategiju sa kompanijom Nokia koja isporučuje i radio i paketno jezgro, u okviru petogodišnjeg 5G plana za pokrivanje glavnih populacionih centara i transportnih koridora, koristeći mrežu tornjeva izdvojene kompanije EuroTeleSites-a kako bi zadržala RAN relativno rasterećenim od fiksne imovine (asset-light).

Telekom je, nasuprot tome, usred zaokreta sa nasleđenog 2G-4G otiska koji se u velikoj meri oslanjao na Huawei, ka dvostrukom nordijskom 5G sloju, finansiranom kroz karakterističnu multilateralnu strukturu koja uključuje Evropsku banku za obnovu i razvoj, švedsku izvozno-kreditnu korporaciju i američku Exim banku. Značajno je da ugovor sa američkom Exim bankom od približno 42,97 miliona evra (50 miliona dolara) predstavlja prvi sporazum te vrste između jednog evropskog telekom operatora i američke Izvozno-uvozne banke. Ericsson i Nokia su ugovoreni za isporuku 5G baznih stanica za uvođenje mreže.

Yettel, preko CETIN Srbija, takođe prolazi kroz aktivnu diversifikaciju dobavljača dok prelazi sa Huawei-a, koji je istorijski korišćen za njegov RAN, na Ericsson za izgradnju 5G mreže. Ova tranzicija se koordiniše uporedo sa širim projektom “5G Balkan” kompanije CETIN, koji ima za cilj jačanje optičke povezanosti širom regiona u partnerstvu sa CETIN Bugarska.

5G podsticaj stiže u pravom trenutku za EXPO 2027

Istovremeno lansiranje 5G mreže u decembru 2025. od strane sva tri operatora, podržano velikodušnim alokacijama od 130 MHz u srednjem opsegu i modernom opremom koja je bila pripremljena tokom regulatornih odlaganja, lansiralo je Srbiju sa bliskog dna balkanske mobilne rang liste na visoko konkurentnu poziciju globalno, potvrđujući potencijal da kasni usvajaoci smanje razliku u odnosu na konkurente kada se usklade politika spektra i realizacija operatora.

Sa EXPO 2027 na horizontu i vladom koja je posvećena korišćenju ovog događaja kao izloga za digitalne sposobnosti zemlje, srpski mobilni operatori suočavaju se i sa pritiskom konkurencije i sa značajnom prilikom da pokažu šta ulaganja u 5G infrastrukturu mogu doneti potrošačima i preduzećima. Rani podaci sugerišu da su dorasli izazovu.

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.