The problem with internet on the UK’s trains isn’t the speed – it’s the wait
Key Points
- Ofcom's 2026 train study found mobile networks meet the "Good Performance" standard less than half the time, with EE leading at 42%.
- Latency, not download or upload speed, is the main reason mobile tests failed on Britain's railways.
- Only EE recorded a median response time fast enough to pass Ofcom's 50ms latency threshold.
- Frequent handovers between masts (up to one every 3.3 seconds) drive the latency spikes that break train connectivity.
- 5G Standalone dramatically reduces latency but is deployed across only 0–32% of journeys depending on operator.
- Peak-time congestion sharply worsens latency, with O2's pass rate dropping 38 points at rush hour.
For the better part of a decade, the mobile industry has trained us to care about one number.
Megabits per second. It is printed on the box, shouted in the adverts, baked into the little badge at the top of your screen that flickers between 4G and 5G as though the distinction were a moral one. Faster, faster, faster. We have been taught to want a wider pipe.
Ofcom’s latest research, carried out by Streetwave across 50 journeys on 24 of Britain’s busiest rail lines, suggests we have been watching the wrong number all along.
The findings make grim reading on their own terms. The best-performing network, EE, met Ofcom’s “Good Performance” bar on just 42% of the journey.
Three managed 21%, O2 20%, Vodafone 17%. On a train in Britain in 2026, the most generous reading is that your phone works properly less than half the time.
But the interesting part isn’t the failure rate. It’s what causes it.
The threshold has three components: a download speed, an upload speed, and a response time of 50 milliseconds or less.
You would assume, given everything we’ve been sold, that the speeds are what trip people up. They aren’t. The study is blunt about it: even when download and upload comfortably cleared the bar, latency was repeatedly the thing that failed.
Of the four operators, only EE posted a median response time quick enough to pass. The other three sat in the low sixties of milliseconds, a hair over the line and a world away from usable.
| Operator | Median latency (ms) | Passes 50ms threshold? |
|---|---|---|
| EE | 46 | Yes |
| O2 | 62 | No |
| Vodafone | 63 | No |
| Three | 64 | No |
| Source: Ofcom / Streetwave, Connectivity on Trains Measurement Study, June 2026. Latency measured as round-trip time. | ||
What is latency?
Latency is the lag. It is the half-second of nothing after you unmute yourself on a video call, the spinner that appears before the page does, the gap between tapping and the world responding. Bandwidth determines how much you can move. Latency determines whether the thing feels alive. And a train, it turns out, is almost purpose-built to ruin it.
Consider what your handset is actually doing as the carriage moves. It is not sitting politely on one mast drinking in signal. It is sprinting through a corridor of them, and every time it crosses a boundary it has to let go of one cell and grab the next.
The study clocked these handovers at a median of four a minute, roughly one every fifteen seconds, and on the worst stretches eighteen a minute, a fresh cell every 3.3 seconds.
Each handover is a tiny stall, a moment where the connection holds its breath while it reconfigures. String enough of them together and the wait becomes the experience.
When signal quality dips, the network re-sends data it couldn’t deliver cleanly the first time, and each retransmission on a 4G connection adds roughly eight milliseconds. A few in a row and you’ve blown the 50ms budget without a single megabit going missing. The pipe was wide. The water just arrived late.
What makes the report quietly hopeful, rather than merely damning, is that the fix already exists and was caught on camera. Where operators have deployed proper 5G Standalone, with no ageing 4G core dragging on the control channel, latency collapses and pass rates jump, in EE’s case to 67%. The trouble is deployment.
5G Standalone covers somewhere between nothing and a third of the journey depending on whose network you’re on. The technology that would fix the problem is sitting in a warehouse marked “rollout pending.”
And then there is the cruelty of timing. The study ran the Bedford to St Pancras line at rush hour and again in the lull. At peak, with the carriage full, O2’s latency pass rate fell by 38 points. The precise moment the most people need their phones to work is the moment the network is least able to oblige.
The megabit was always a vanity metric, a number that looks impressive on a billing page and tells you almost nothing about whether your morning commute will let you join the standup. The thing you actually feel, the thing that decides whether the connection is real or theatre, is the wait.
The networks now have to decide whether they’d rather keep selling speed, or start delivering the one number that counts.
