This is how Britain wins the AI race

Docking at the King Charles III Space Station this week is James Wise: partner at Balderton Capital, author of Startup Century, and now the chair of the government’s Sovereign AI unit — a man who has spent twelve years giving exceptional people money to build companies and has now taken a tour of duty to do something stranger, which is to make the British state behave a little more like an investor and a little less like a grant machine. Tom and Calum want to know the obvious thing first: inside government, is anyone freaking out enough about AI? The good news, James says, is that most people are freaking out less than he is. That “British calm” is reassuring in a crisis and maddening in a transformation, and the real task is not getting one department to understand AI but getting the idea to diffuse across the whole of Whitehall the way a technology diffuses across an economy. Direction of travel is good. “But it’s a vector. We probably need to increase the speed.”

Sovereignty, James argues, is not autarky — it is leverage, and it comes in three levels. There is resilience, the floor, which simply means nobody can turn off your chips overnight: you don’t need British-only silicon, you need three or four suppliers. There is economic sovereignty, where Britain and ordinary Britons actually capture some of the upside — which is why he pushes back on DeepMind as a simple tale of loss, given the income taxes, the retained team, and the billions of Alphabet’s money that built it here. And there is the full-stack version, the flag on the building, legal and regulatory alignment with the British state. The naive dream — every electron and every wafer made between Somerset, Swindon and Glasgow — collapses on contact with the facts: a single token generated by a GPT model draws on something like 35 countries, ASML’s lithography machines are atom-level miracles the Americans and Chinese are pouring unlimited money into merely replicating, and the quartz comes from one mine in North Carolina. Doing the whole stack ourselves would be “beyond a Manhattan project” — it “would make Bletchley Park look like a weekend crossword.” ARM is the lesson: best in the world, and still unable to make a single chip alone. Sovereignty “is not end to end. It is having enough power to mean no one can turn you off.”

SovAI runs like a venture fund — taxpayer takes the loss, taxpayer keeps the upside — and the criticisms come from both sides at once: that the state should stay out entirely, and that £500m is far too small to matter. James thinks both are partly fair. But the deeper problem he keeps returning to is capital. Anthropic just raised $65bn in a single round; no institution in Britain, and barely any in Europe, could write that cheque. A faltering London Stock Exchange means our best companies IPO abroad, and because UK pension funds have been driven out of equities, Britain’s venture scene quietly “feeds promising companies to the US.” The culprit, on his telling, is a well-meaning fee cap that treated a decade of patient, board-level company-building exactly like a hedge fund pressing a button to buy a share. The human cost is concrete: a coal-miners’ pension fund and a supermarket-workers’ pension fund put one or two percent into his early fund as a last hurrah before winding down, rode Revolut, GoCardless and Depop, and left their members thousands — maybe tens of thousands — of pounds better off. Since then, Canadians and Danes and Americans have shared in Britain’s winners. Very few Brits have. “And it’s really sad that that’s happened.”

James is an avowed Schumpeterian who thinks the model-builders themselves are catastrophically bad at describing a good future, too “obsessed with the dystopian.” When Calum raises the spectre of Schumpeter’s tornado — what if all cognitive and manual work is automated? — James calls it an extremist argument: the tractor once did the work of fifty men, a calculator already beats you at maths, and the countries that navigated those shifts well saw gains never before seen in human history. The most tangible place Britain will feel the upside, he insists, is healthcare. Most of medicine has been “getting better at guessing”; now we have the best predictive model of human health ever built, able to target the drug to the cancer killing your loved one — and serious parties polling in double digits want to keep it out of the NHS, which is like throwing out the calculators and the blood tests. He reaches for Florence Nightingale, who brought data visualisation to the Crimea and had her own dissenters, as the British template: the scientific entrepreneur. The NHS, the largest single-goal workforce on Earth, is the asset. Forget that we have no NVIDIA — “what happens if we had AstraZeneca 2.0, or next-generation Nightingale?” And Jensen Huang, getting treated here for his cough, “is not going to turn the chips off.”

Most things, James believes, are downstream of culture, and Britain’s culture got caught in a post-war “pincer movement” that offered two respectable lives — big state or big corporate, a job for life and a predictable mortgage, the Volvo moving slowly up the cul-de-sac. That settlement is over: automation thins the big employers, “the maths doesn’t math anymore on a massive state,” and the way out is entrepreneurship — except the whole system, from university debt to mortgage underwriting, is rigged toward the salaried path. He wants a Britain where more people are willing “to build, to ship, to fail, to rebuild,” with risk capital available to anyone. He is genuinely heartened by the fourteen-year-olds running Depop stores and disheartened by a youth politics that sees opportunity only in taking from others — “not a very British” solution. And he ends on an idea from Startup Century: since half of all student loans are never fully repaid, the state is already in the high-risk-capital business, so offer the same money as a no-recourse loan to start a company — “which is venture capital, by the way.” Want to do an English degree? Great. Want to explore a new use of radar for cricket? Here’s £50k. Have a go.

The episode explores

— What the Sovereign AI unit actually is: a state venture fund that takes the loss and keeps the upside, not a grant-and-forget machine

— Why getting AI right means diffusing the idea across all of Whitehall, not briefing one department — “it’s a vector; we need to increase the speed”

— The three levels of sovereignty: resilience, economic benefit, and full-stack legal alignment — and why “we can’t have someone turn off our chips overnight”

— Why DeepMind is not a simple story of loss when you count the taxes, the retained team, and Alphabet’s billions

— The single GPT token that needs 35 countries, ASML’s near-impossible machines, and the one North Carolina quartz mine

— Why doing the whole stack ourselves “would make Bletchley Park look like a weekend crossword,” and why ARM proves sovereignty is leverage, not end-to-end control

— The capital bottleneck: a $65bn Anthropic round, a faltering LSE, and pension rules that make Britain feed its best companies to America

— The fee cap that treated a decade of company-building like a hedge fund pressing a button — and the coal-miners’ pension fund that rode Revolut and Depop

— Why the people building AI are the worst at describing a good future, and what Schumpeter does and doesn’t tell us

— Most of medicine has been guessing — and why keeping AI out of the NHS is like throwing out the calculators

— “Most things are downstream of culture”: the post-war pincer, and how to make Britain entrepreneurial again

James Wise is the author of Startup Century (available from Amazon, Waterstones and others; the audiobook, he notes ruefully, was read by an American) and a partner at Balderton Capital. Read about the government’s work at the Sovereign AI website.