The pandemic exposed a huge weakness in cable company networks when it became clear that their networks did not have enough upload capacity to support people working and schooling from home. That period when people struggled to work from home might have been the trigger to convince millions of people that fiber was superior to cable technology. The cable companies reacted quickly and goosed upload speeds to the range of 30-40 Mbps. Since then, they have slowly been upgrading to much faster upload speeds using mid-splits and DOCSIS 4.0.
A recent article from Ookla suggests that the same need for faster upload speeds might be coming for cellular networks due to the way that people are starting to use AI in daily life. The article provides some examples of ways we might use AI in the near future. A person might scan a menu in a restaurant, and AI can provide real-time feedback to estimate the calories in each dish or highlight foods that might trigger an allergic reaction. This would require quickly uploading a picture of the menu to provide quick feedback. That’s not a data-intensive transaction, but consider instead using AI to provide real-time feedback to somebody walking around in a foreign city. AI could translate signs and describe the nature of stores or shops as they come into view.
U.S. cellular companies have allocated the smallest percentage of bandwidth to upload compared to the major cell providers around the world. AT&T, T-Mobile, and Verizon have allocated between 6.6% and 7.1% of total bandwidth capacity to upload. In contrast, China Telecom and China Unicom have allocated over 16% of bandwidth to upload.
In writing this blog, I took a speed test on AT&T and got a speed of 381/11 Mbps on my cellphone. I note this is the fastest download speed I’ve ever received on AT&T, by a lot, and shows the impact of the AT&T’s recent introduction of the spectrum acquired from EchoStar. I took several other tests with similar results, and at my house, the upload speeds are only about 3% of total bandwidth.
American cellular carriers seem to be in a race to claim the fastest network for bragging rights, and this has led them to maximize download speeds to an extreme degree. I doubt that many people are complaining except for folks who are trying to stream video from their phone. When I swap my phone over to WiFi, the upload speed in my Charter connection is more than 10 times faster than the AT&T cellular upload connection.
The article points out that carriers have options to boost upload speeds. The one that is discussed the most in the article is to convert cellular networks to dynamic TDD (time division duplexing), which would allow the phone to assign bandwidth available to the phone to either download or upload, according to the immediate need.
But that fix alone wouldn’t solve the problem, because a carrier would need to beef up the entire network in the upload direction to handle larger volumes of uploaded data. There are other interesting limitations. For example, if a carrier uses shared spectrum like CBRS for uploading, then setting a faster upload would have to be coordinated with the other major users of the spectrum to synchronize the network clocks.
The Ookla article also demonstrates that handsets can be a limitation by showing the upload speeds that can be achieved on different generations of Samsung Galaxy phones. with lower upload capability on older phones.
The slow upload speed on my tests might be an anomaly, but before AT&T introduced the new spectrum, my upload speeds were rarely faster than 5 Mbps. Ookla says that median upload speeds in the second half of 2025 were 18 Mbps for AT&T, 21 Mbps for Verizon, and 27 Mbps for T-Mobile – all slow in comparison to fiber and upgraded cable technologies.
The AI people are using reduces upload needs, less local resources being used and primarily text prompts or individual image uploads.
Increases in upload use are almost entirely driven by syncing of data like icloud or google device backups etc.
The smart glasses with video streams is about the only thing visible on the horizon, and frankly the latency to upload video, process, and feed back to users I think makes this unlikes. Every company with an AR glasses system is doing local AI compute, so this is somewhat unlikely to drive upload demand.
The biggest practical realtime upload demands have already come, and that’s your zoom/teams/facetime calls.
The problem is that we’re still in a bandwidth race but the need is lower latency, lower jitter, more reliability. As little as 50×10 is suitable for a home, so if we can do 250×25 to a mobile device, home, etc then we’re WAY into the comfortable zone for essentially everything if the latency and losses are good (they aren’t). People are buying ‘faster’ plans thinking it solves their latency and loss issues and it doesn’t.
Yeah, I would say 98% of the request for faster speeds from our clients dissolves down to they have some latency/jitter issue they are trying to fix. Once that problem is cleared up, their existing speed is again 100% satisfactory.