“I just feel like there’s a slightly dishonest debate in this country”, Clegg told City AM. “No one in their right mind would ever train an LLM foundation model in the UK. In fact no one does”.

“I think given our relatively marginal relevance and importance as a country in this great battle between China and the US, we all have to slightly grow up about about what we mean about being completely British”. Just days ago, the government launched its £500m Sovereign AI venture fund, promising homegrown startups and early-stage investment, access to the country’s supercomputers, as well as fast-track visas for research talent. Tech secretary Liz Kendall dubbed it “unlike anything government has ever done before”.

‘We’ve skipped it altogether’ The reason the UK never built its own foundation models is, according to Clegg, a direct consequence of policy choices. Specifically, the decision to side with rights holders over AI developers on copyright. “Energy is too expensive”, he told City AM.

“The IP copyright stuff – it’s a perfectly logical decision for a government to side with content providers, whether its [the media], the publishers or whether its Elton John or Paul McCartney.” “But if you do that”, he added, “you’re never going to attract foundation model LLM developers”. The result, according to Clegg, is a country that stumbled out of the AI industrial revolution before it began. “We are now operating as a country without a single steam engine we can call our own”, he told City AM.

“That’s why we’re dependent on US tech – by accident or by design, we’ve just skipped that altogether”. He revealed he attempted to make the case directly to the prime minister and the chancellor. “The last time I ever spoke to Keir Starmer and Rachel Reeves about this a year ago, I tried, and I think I singularly failed to get them interested”, he told City AM.

The broader sovereignty concern, he stressed, is real and serious, even if the current debate around it is confused: “It’s an astonishing level of over-reliance on American tech”. ‘It’s done. So we now need to decide what’s next’ Instead, Clegg’s pitch is to lean into open source models, which can run on domestic infrastructure with no data link back to an American server.

“The leading open source models are now all Chinese,” he said. “I personally would not be particularly squeamish about using open source Chinese models. “They might not in their crude state spit out a particularly coherent response about what happened in Tiananmen, but you could run them in a sovereign way, you can run them free, and you can customise them exactly as you want.” Clegg’s longer-term optimism rests not on large language models but on what he believes comes next.

He is an investor in a post-LLM research lab being established in Paris by Yann LeCun, and recently joined the board of Nscale, the British data centre developer valued at $14.6bn following a $2bn fundraise backed by Nvidia. The company has pledged to build what it calls the largest UK sovereign AI data centre in Loughton, Essex. His wager then is on so-called world models, which understand the physical world rather than processing language. “You’d use an LLM to talk to your domestic robot – you would never use it to get your robot to walk around the house, it would smash up the furniture”, he told City AM.