Emerging AI technologies could put an increased focus on mobile users’ upload connections and capabilities.
Most mobile users today are mostly concerned with surfing the web, watching videos and checking social media – all tasks that center on their phone’s ability to pull data down from the internet.
But this may change as AI sails into the mainstream. Already services like ChatGPT can analyze users’ uploaded pictures. And some smart glasses hint at a future where AI could provide real-time insights into an uninterrupted, uploaded live stream of users’ daily activities.
This kind of future could put new demands on mobile networks. Thus, it’s worth looking at how mobile network operators globally have managed their uplink connections to date, in the shadow of this possible AI future.
Key takeaways:
- Of the 17 major operators analyzed in Ookla Speedtest Intelligence® data, U.S. providers allocated the smallest percentage of capacity to users’ uplink connections. Chinese operators allocated the largest percentage.
- Mobile upload speeds have been rising globally from 2021 to 2025, thanks to the release of additional spectrum and a variety of technological advances. But operators have not been increasing the percent of network capacity allocated to uplink connections during this period. Some have been reducing that percentage.
- According to Ookla RootMetrics® drive test data for the U.S. market, T-Mobile, AT&T, and Verizon all allocated 20% of their Time Division Duplex (TDD) midband network “slots” to uplink connections in the second half of 2025. TDD is widely used among 5G network operators to determine the ratio between uplink and downlink resources in midband spectrum.
- Future AI technologies could create new networking demands. For example, widespread adoption of smart glasses – those that upload users’ live views for AI analysis – may create a multitude of lengthy, continuous video streams that could pose difficulties for bandwidth-constrained uplink connections. At the same time, however, there are plenty of unknowns in this emerging space. For example, will most future AI requests be answered by software running inside of users’ phones, thus eliminating the need for a network connection entirely?
Anticipating the effect of AI on the network
Meta’s newest smart glasses allow users to receive an AI analysis of what they see. Google’s Gemini Live provides a similar service. These types of offerings hint at a new paradigm of computing that could eventually stream users’ live video feeds directly to an AI analysis bot. This constant visual data stream could even allow AI to proactively understand a user’s context in real-time and offer immediate, situation-specific assistance, without being prompted.
Real-world scenarios using this type of technology abound. For example, as a user looks at a broken appliance, an AI bot could identify the specific model, access repair manuals, and highlight exactly which component needs attention via audio and visual cues. It could provide instant translation of foreign street signs as a user walks past them, or it could offer nutritional analysis of food via a glance at a menu.
If these kinds of services become popular, high levels of uplink traffic could put additional demand on mobile networks globally. Already 5G equipment vendor Ericsson has speculated on what this AI future might mean for mobile network operators. “The uplink traffic will increase significantly over the coming years and, indeed, is becoming telecom’s new ‘currency,’” the company wrote. “This potential growth of uplink traffic underlines the importance of network capacity planning, spectrum allocation, and RAN [radio access network] feature developments.”
Ericsson isn’t alone. “AI changes how traffic is generated, where it flows and when it peaks. It increases uplink use in the home, it injects automation and machine vision into industrial sites, and it multiplies east–west movement between data centers,” Nokia wrote in its own report on future mobile traffic, including from AI.
And in a new report, the GSMA trade association offered three different scenarios for future growth of traffic on mobile networks globally. “In the low-growth scenario, the downlink remains predominant at around 85% of total traffic, with uplink at 15%,” the firm wrote. “However, in the medium- and high-growth scenarios, the share of uplink increases to around 25% and 35%, respectively, by 2040.” The reason? AI.
But such predictions are just that: forecasts that may – or may not – come true. There is still much uncertainty regarding the parameters and the extent of AI traffic on a wireless network. For example, it’s not clear how much AI processing will ultimately be conducted on users’ devices and how much will need to be routed through a network connection to a cloud-based computing service. This question is central to forecasting AI’s eventual networking demands.
Another unknown involves the speed at which AI requests will need to be answered. This topic sits in the realm of latency – the time it takes for a cloud-based service to respond to a request from a user – and it too will have serious implications for eventual AI networking designs.
Yet another unknown involves the extent to which 5G will play a role in an AI future. How many AI requests will be routed through wired and Wi-Fi connections? Will those types of non-cellular networking connections be pervasive enough to reduce any possible AI strain on a 5G operator’s network?
Such questions go on and on: Will newer video compression technologies ease upload bandwidth demands? Will technological advancements – such as those from new 5G-Advanced standards – make wireless networks even more speedy and efficient?
With all that said, now is the time for 5G operators to begin considering how AI might affect the usage of their networks. For years now, mobile users globally have been sucking down movies and memes on the downlink. But AI may flip that: It will need eyes and ears to work, and that could translate into massive streams of data flowing up from users to the internet.
Operators gauge uplink capacity allocations
Frequency Division Duplexing (FDD) was used widely in previous generations of cellular technology, including 4G LTE, to determine the amount of capacity allocated to users’ uplink connections. FDD divides users’ uplink and downlink connections into two dedicated, separate channels. Think of FDD traffic as lanes on a highway separated by a concrete barrier: One lane is permanently dedicated to uplink traffic, and the other is permanently dedicated to downlink traffic. This setup worked well for initial networking priorities focused on voice and coverage in spectrum allocations between 5 MHz and 20 MHz.
But FDD doesn’t cut it in the age of 5G. Operators need speedy, efficient and flexible data connections, particularly when they’re dealing with chunks of midband spectrum that can range up to 100 MHz or higher.
“To increase flexibility as well as make spectrum usage more efficient, Time Division Duplex (TDD) is becoming increasingly common and important,” noted the GSMA global trade association.
As a result, many of today’s midband 5G networks use TDD. For example, according to Ookla’s RootMetrics drive testing data in the U.S., a large portion of T-Mobile’s 5G samples used TDD technology. Specifically, around 93% of T-Mobile’s downlink samples used standalone (SA) 5G in the second half of 2025. Of that 93%, 78.6% used TDD and the rest (14.4%) used FDD.
TDD essentially uses a single “lane” for data traffic in both directions, upstream and downstream, but it rapidly switches the direction of the flow of traffic thousands of times per second. By adjusting the timing of this “traffic light,” an operator can decide to keep the light green for downloads for 80% of the time and only switch to uploads for 20% of the time, for example.
While dynamic TDD allocations may be possible – where an operator adjusts its traffic light in real time to meet a sudden surge in users’ uplink demands – most operators stick to static, synchronized patterns to maintain network stability and prevent interference.
Upload capacity varies by operator and geography
Since operators have some control over the network resources they allocate to uplink connections versus downlink connections, it’s worth looking at how some of the biggest operators in the world handle this decision.
To do so, we used the relationship between upload and download speeds as a basic proxy for carriers’ allocation of networking resources toward uplink and downlink capacity. (This is separate from Ookla’s Speed Score® that incorporates download speeds, upload speeds, and latency).
To be clear, upload and download speeds are a product of operators’ network capacity decisions, but they can be affected by a wide variety of factors including operators’ spectrum holdings and the capabilities of users’ phones. Nonetheless, this study of operators’ upload and download speeds helps to shine a light into their capacity priorities within the parameters of what they can control.
The below findings are from Ookla Speedtest Intelligence data. They show the percent of networking capacity dedicated to operators’ uplink connections, which we calculated by dividing aggregated upload speeds by the sum of operators’ aggregated download and upload speeds. We only used the top 10% fastest 5G download samples (using both FDD and TDD) in order to obtain a clearer view into operators’ networking designs, one that’s unimpeded by connections potentially suffering from interference, network congestion, or other issues.
Of the operators studied, it’s clear that Chinese wireless network operators like China Telecom and China Unicom likely allocate a larger portion of their network capacity to users’ uplink connections
On the other end of the chart, U.S. operators like Verizon, AT&T, and T-Mobile likely allocate a smaller portion of their network capacity to users’ uplink connections.
It’s also worth noting that overall spectrum ownership doesn’t appear to affect operators’ approaches to uplink capacity. For example, according to the GSMA’s Spectrum Navigator, Vodafone holds the most overall midband and lowband spectrum (526 MHz) of the 17 operators studied. China Telecom sits in the middle of the pack with 220 MHz of midband and lowband spectrum. And AT&T sits near the bottom with 172 MHz.
The ratio between uplinks and downlinks could reflect a wide range of factors such as differences in customers’ usage patterns, device capabilities, local competition among operators for the title of fastest provider, network vendor capabilities, and other parameters. Indeed, some Asian network operators have been highlighting service plans that focus on uplink performance as a way to entice livestreamers.
That focus on the uplink could expand to other markets. Uplink is “one of most under-talked topics of the industry,” T-Mobile Chief Network Officer Ankur Kapoor recently told Fierce Network.
As speeds rise, downlinks dominate
Most users around the world are enjoying faster uplink speeds than ever before. For example, overall mobile upload speeds in the U.K. increased by around 36% between 2021 and 2025, according to Ookla Speedtest Intelligence data for the market’s 10% fastest connections across all mobile technologies. In the U.S., that figure is around 40%. Other leading 5G markets have seen similar improvements.
This uplift can be traced to a wide variety of factors ranging from additional spectrum allocations (more spectrum typically results in faster speeds) to technological innovations like carrier aggregation (which can speed up connections by “glueing” together transmissions in different spectrum bands) and MIMO antennas (which can transmit and receive simultaneous data streams).
But there are also plenty of caveats. Yes, mobile upload speeds have been rising globally, but that’s mainly because 5G enables faster overall connections, both on the uplink and the downlink. In some countries, like Brazil, the percentage of network capacity allocated to upload speeds has been falling. In other countries, like China, the capacity allocated to the uplink has been holding relatively steady. In no country in this study is the percentage of capacity allocated to the uplink rising in a significant way.
Percent Capacity Allocated to Upload
Speedtest Intelligence | 2021 – 2025
The data reveals a clear trend: as mobile network technologies mature and meet baseline user needs for upload capacity, operator focus pivots toward driving ever-faster download speeds. Indeed, overall mobile download speeds in the U.K. increased by around 58% between 2021 and 2025, according to Ookla Speedtest Intelligence data for the market’s 10% fastest connections across all mobile technologies. In the U.S., that figure is around 251%.
Thus, while uplink performance is improving, the proportional importance of download capacity continues to dominate operator investment and network configuration choices.
Here too are caveats. Network operators can tweak their networks in different ways for different locations. For example, venues like sports stadiums or concert halls may feature network settings and equipment tuned in ways that aid fans’ uplink connections.
Another important caveat: Nationwide standards for TDD connections designed to prevent interference. As explained by the GSMA, all the operators in a given geographic area that use 5G TDD in spectrum bands like 3.5 GHz must synchronize their network “clock” and frame structure so that all their users transmit and receive data at the exact same times. This helps prevent interference and avoids large, inefficient physical separation zones between networks. This may explain the similarities in uplink percentages among operators in the same geographic markets in recent years.
RootMetrics highlights operators’ uplink settings
Aggregated Speedtest results are one way to gain a view into operators’ uplink calculations. Another, more exacting method is via RootMetrics drive test results. Such tests – using flagship, off-the-shelf Android smartphones – provide a deeper look into operators’ network settings via 11 million total tests conducted annually. RootMetrics administers controlled, nationwide testing in the U.S. and elsewhere.
A sampling of RootMetrics’ insights into uplink connections: Roughly 79% of T-Mobile’s uplink sample tests in the second half of 2025 traveled over the operator’s 2.5 GHz midband spectrum holdings. And just over one-third of those samples used two-carrier aggregation technology. By aggregating multiple carriers on the uplink and downlink, operators can increase users’ overall speeds.
More importantly, RootMetrics data can offer a closer look at the uplink settings deployed by U.S. wireless network operators. For example, it can show the number of network “slots” allocated to uplink connections. In TDD networks, “slots” are the specific time intervals – typically measured in milliseconds – within a transmission frame that are designated for either sending data (downlink) or receiving data (uplink).
According to RootMetrics U.S. data, T-Mobile, AT&T, and Verizon all allocated 20% of their TDD midband network “slots” to uplink connections in the second half of 2025. However, the operators’ median upload speeds during that period showed some variation:
There are a variety of reasons behind these results. For example, T-Mobile was an early mover to 5G standalone (SA) technology, which generally supports speedier connections than non-standalone (NSA) architecture. Furthermore, T-Mobile’s midband 5G network sits in 2.5 GHz spectrum, whereas Verizon’s uses C-band and AT&T uses both 3.45 GHz and C-band. And the operators also apply different levels of carrier aggregation to their uplink connections.
That last item – carrier aggregation – can have clear impacts on users’ uplink speeds. For example, the Samsung Galaxy S24 and S25 Ultra smartphones are endowed with carrier aggregation technology for uplink connections. Such technology binds two bands of spectrum together to improve network capacity and upload speed. T-Mobile in 2024 enabled two-carrier aggregation for uplink connections on its 5G SA network. The result of this deployment can be seen within Speedtest data:
Upload Speeds Among T-Mobile's Top 10% Fastest 5G Samples
Speedtest Intelligence | Q4 2025
In RootMetrics’ testing in the fourth quarter of 2025, roughly 37.8% of T-Mobile’s 5G SA samples in its 2.5 GHz spectrum used two-carrier aggregation technology.
Ookla analyst Kerry Baker contributed to this article.
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.
