King’s Cross AI boom reopens Britain’s ‘sovereignty’ debate

TL;DR:

• A cluster of US AI giants — Google, Meta, OpenAI and Anthropic — has turned London’s King’s Cross into Europe’s densest AI district, much of it traceable to DeepMind staying in the city after its 2014 Google sale.
• Washington’s abrupt order barring foreign nationals from Anthropic’s latest models has revived a decade-old argument over UK “AI sovereignty”.
• Ministers are backing a spread of smaller bets — a £500m Sovereign AI fund and up to £60m for Oxford and UCL labs — rather than a homegrown frontier model.

Two decades after King’s Cross was one of central London’s most neglected quarters, it now houses the main European bases of several of the world’s wealthiest firms. The turnaround traces largely to Sir Demis Hassabis, who kept his DeepMind lab in London after selling it to Google in 2014 rather than relocating to America’s West Coast.

That history has taken on fresh weight. With barely 90 minutes’ notice, the US Department of Commerce told Anthropic to stop any foreign nationals using its frontier models, Fable and Mythos. As Air Street Capital’s Nathan Benaich framed it, the most advanced AI is built by a handful of American firms under American law, and what others may do with it “can change on a Friday afternoon”. The episode pushed sovereignty onto the agenda at this month’s G7.

The build-British question

The so-called “DeepMind mafia” has spun out richly funded start-ups such as Ineffable Intelligence and Recursive Superintelligence — yet none is building the kind of large language model UK firms and departments could lean on independently. Of the $55bn raised worldwide by DeepMind alumni, only around $5bn has landed in Britain. Structural gaps compound the problem: scarce compute, a sluggish planning system and grid constraints that recently sent SoftBank’s investment to France and left OpenAI’s “Stargate UK” on indefinite hold.

Westminster’s measured bet

AI minister Kanishka Narayan is sticking with a portfolio: the £500m Sovereign AI fund, £120m in grants, and up to £60m for “blue sky” labs at Oxford and UCL, which we covered when they were announced this week. One contender, Cosine, is using Bristol’s £225m Isambard-AI supercomputer to train what it claims will be Britain’s first sovereign frontier model by year-end. The push echoes the wider scramble after the US curbs, from European firms spreading their AI bets to a legal-tech firm suing Washington over the Anthropic order.

Looking forward

Narayan’s wager is that Britain should “bet on where the future is going” — efficiency and new paradigms — rather than refight the scaling race it cannot fund. Whether that counts as sovereignty, or simply a polite acceptance of dependence on US labs, is the question the next government inherits.