British researchers have pushed the boundaries of internet performance, delivering a breakthrough that could reshape how data moves through existing networks.
A team at University College London, led by Polina Bayvel, has achieved a data transmission speed of 450 terabits per second. To put that into perspective, the capacity matches the simultaneous streaming of roughly 50 million films.
What makes this milestone stand out isn’t just the raw speed—it’s how the team achieved it. Researchers transmitted data between a lab in Bloomsbury and a data centre in Canary Wharf using fibreoptic cables already installed beneath London’s streets. No new infrastructure. No disruptive upgrades.
That approach delivers a striking implication: networks could expand dramatically without the cost and complexity of laying new cables. The improvement represents a tenfold increase over current commercial capabilities. In practical terms, it’s comparable to adding nine extra cables for every one already in place—without digging up a single road.
The timing matters. As artificial intelligence systems scale, they demand ever-larger volumes of data. Current AI workloads operate in the range of 100 to 400 gigabits per second, and that figure continues to climb. The question isn’t whether demand will grow, but whether infrastructure can keep pace.
Dr Bayvel captured the tension between human limits and machine demand:
“There’s only a certain amount of data that anyone could process – you can only watch so many movies.
“But AI infrastructure is generating a lot of data, and that data is is spewing into the network.”
To unlock these speeds, the team engineered custom hardware capable of transmitting data across a broader spectrum of light frequencies than standard commercial systems. They extended the usable range from 1264 to 1617.8 nanometres, then developed new methods to manage distortion across that expanded bandwidth.
While laboratory experiments have previously achieved faster speeds under tightly controlled conditions, this test marks a shift toward real-world application. It proves that existing infrastructure can handle far greater capacity than previously assumed.
The path to deployment now looks shorter than expected. Researchers suggest the technology could reach wider implementation within five years.
Industry observers see immediate value. Kerrianne Harrington highlighted the practical advantage of building on what already exists rather than starting from scratch:
“The interesting thing about this work is it’s using what’s already in the ground, which is the expensive thing to change.
“I do think it’s a very practical approach to the problem. I would say that the work that’s shown in this paper has a more immediate benefit to increasing capacity than new fibres.”
The breakthrough reframes a familiar challenge. Instead of asking how to build more infrastructure, the focus shifts to how much more performance current systems can deliver.
If networks can scale this way, the next phase of digital growth—from AI to streaming to cloud computing—may depend less on construction and more on innovation within the cables already beneath our feet.
Author: George Nathan Dulnuan