That's not a metaphor. It's a literal, specific choice, and a fortnight ago I'd have told you Britain was choosing the wrong side of it by default, by saying nothing at all.
Then the government said quite a lot, very quickly, and I need to deal with that before anything else.
On the 20th of July, Andy Burnham stood on the steps of Downing Street and gave his first speech as Prime Minister. He talked about "reindustrialising Britain, using public procurement to back British industry." He promised a government that would be "a circuit breaker for Britain, bringing forward the biggest changes in the last forty years." He talked about a ten-year plan, due later this year. He did not say the words AI or artificial intelligence once.
I wrote at the time that the silence was the story. It wasn't, and the correction matters, because what replaced it is more damning than the thing I was complaining about.
Later that same day, Burnham made Kanishka Narayan the first Minister for Artificial Intelligence in British history to sit at the Cabinet table. He abolished DSIT outright, splitting its work across a new Department for Business, Innovation, Science and Trade, DCMS, and a new Office for the Prime Minister and the Cabinet. Four days after that he put Lord Vallance in charge of an AI Taskforce reporting directly to him, and folded the AI Security Institute into it, so that strategy, safety and industrial policy now sit under one roof at the centre of government.
Narayan's own framing was almost exactly the thing I said was missing: "AI is likely the most significant technology in human history. That is why it is now at the UK Cabinet's table, so we can reindustrialise with AI, secure our country, and support our public services."
Credit where it's due. In five days AI went from unmentioned to the most centralised policy area in Whitehall. Nobody I know predicted that, including me.
Here's the revised complaint. They rebuilt the organisation chart in four days. They have not moved a single pound.
Go through the money and check the dates.
The £500 million Sovereign AI Fund, running like a venture arm inside government with equity cheques and compute time attached, announced under Starmer and paying out its first cohort in April. The £1.1 billion AI Hardware Plan, announced by Liz Kendall at London Tech Week on the 8th of June, with £750 million of it going to a national AI supercomputer at Edinburgh due to land in 2030, £400 million for next-generation chips and £120 million for a hardware innovation programme. The £25 billion package from last November, which was mostly private capital from Vantage, Microsoft and Equinix rather than Treasury money. Our sovereign compute plan, the AI Research Resource, went from 2 to 21 AI ExaFLOPs last year, a real tenfold increase, with a target of 420 by 2030.
Add up every scheme going and total government commitment to sovereign AI comes to around £3.5 billion. That number has not moved since Burnham walked through the door. He has added a minister, a taskforce, a chair and a reporting line. Not a pound.
For scale, on the 30th of July, ten days into the job, he announced £8.4 billion for nuclear submarines with 22,000 apprenticeships attached. That may well be the right call. I'm not going to pretend to know more about the deterrent than the people who briefed him on it. But £8.4 billion in a single defence announcement, against £3.5 billion in total, ever, across every scheme, for the technology he says is going to reindustrialise the country, tells you where this actually sits in the ranking.
Three and a half billion pounds against America's half a trillion for Stargate alone. We're not slightly behind. We are playing an entirely different game, at an entirely different scale, and calling it the same competition.
Here's the statistic I used a fortnight ago. In 2025, global AI venture capital hit $258.7 billion, according to the OECD. The US took 75% of it, or $194 billion. The UK took 5%, or $13.8 billion. We came fourth, just ahead of China's 5%, essentially tied with the EU's 6%.
Now here's what happened next, and it's genuinely good news until it isn't.
British startups raised about $17 billion in the first half of 2026, more than double the same period last year. AI took $12.6 billion of that, roughly 74p in every venture pound deployed in this country. We pulled in 39% of all European venture capital, more than France, Germany and Sweden combined, with four separate rounds above a billion dollars. On any domestic measure that is the best six months British technology has ever had. Isomorphic Labs, Nscale, Wayve. Real companies, real money, built here.
Now the other column. Global venture funding over the same six months hit $510 billion, more than the whole of 2025 put together. OpenAI and Anthropic between them took $217 billion of it. Two companies. Forty-three per cent of every venture dollar deployed anywhere on earth, in six months, split between two firms in California.
Put those side by side and sit with it for a second. Every AI company in Britain, in the best half-year the sector has ever had, raised about a seventeenth of what two American companies raised between them.
We doubled our money and lost ground. That's the part I can't get past, and it's a harder problem than the one I thought we had. When you're behind and standing still, you can tell yourself the answer is to start running. We started running. The gap grew anyway, because the thing we're chasing is accelerating faster than we are, and no amount of British effort changes the arithmetic of two firms hoovering up nearly half the world's venture capital.
On the 12th of June the US Commerce Department wrote to Anthropic and placed export controls on its two most capable models, Fable 5 and Mythos 5, citing national security. The order barred access by foreign nationals, including Anthropic's own non-American employees. The company could not sort its users by nationality at scale without trampling their privacy, so it switched both models off for everybody. Around the same time OpenAI was told to limit its newest model to vetted partners.
Eighteen days later Commerce withdrew the controls, and Fable 5 came back globally on the 1st of July.
Read the terms of that ending, though, because it's where the real story is. Mythos 5 did not come back for the rest of us. It was restored for a vetted set of American organisations. Howard Lutnick's letter credited Anthropic's commitments to detect and address security risks, and said plainly that Commerce could impose the licence requirement again if the company fell short. Nothing structural changed. A foreign government switched the tools off, then switched most of them back on, on its own timetable, having extracted commitments in exchange, and reserved the right to do it all again.
That is not a resolved dependency. It's a licensed one.
Whatever workflows, products or advantage British businesses and government departments had built on those models simply stopped working for eighteen days, by decision of a foreign government, with zero say from anyone in London. That is what dependency looks like when it stops being a policy paper and becomes a Tuesday morning where your tools don't work.
Chatham House said it plainly the day Burnham took over: "The UK's dependencies on US technology run extremely deep… a strategy resting on proximity to the US alone looks less sensible with every passing day." The Tony Blair Institute has warned that without meaningfully more compute by 2030, Britain risks holding "less than 1 per cent of the world's total AI compute," down from a best case of around 3%, against America's roughly 75%.
To their credit, this seems to be exactly what's rattling Burnham's team. Reporting from the Financial Times suggests his advisers, including former tech minister Josh Simons, think close alignment with US tech giants under Peter Mandelson's old playbook has been "a geopolitical failure that hasn't delivered on its intended aims." One person involved called it "unfettered tech boosterism."
They're right that the old approach hasn't delivered. Where I think they're at risk of getting it wrong is the conclusion. The answer to "we're too dependent on America" isn't to regulate our way out of the room. It's to build enough of our own capability that we're not simply taking whatever access we're given. Control without capability is just a different flavour of dependency, one where we've made ourselves smaller instead of stronger.
Here's where I have to be honest, because pretending the other side of this argument doesn't exist would make the whole piece worthless. And since I last wrote about it, the other side has got considerably stronger.
A report from Global Action Plan in April found that 84% of proposed UK data centre developments sit in areas already water-stressed, or projected to be by 2040. Slough has at least 32 data centres running in a region the Environment Agency classifies as seriously water stressed, and Thames Water estimates that under a high-growth scenario data centres could drive nearly 30% of all new water demand in the area, around 270 million litres a day. Water UK told MPs that data centres are excluded from the Environment Agency's national water resources framework altogether, and that the AI Growth Zone policy makes no mention of water at all. The national policy statement meant to settle all this was due in February. It still hasn't appeared.
In Scotland there are roughly 24 proposed hyperscale data centres which, if all built out, some estimates suggest could consume up to one and a half times Scotland's peak electricity demand. Edinburgh refused a hyperscale site outright in February. In July the SNP's national council backed considering a national moratorium on new data centres, and with the Greens they have the numbers to pass one.
Then there's the bit that should worry anyone making my argument. A Guardian investigation found that the developer and the UK government misrepresented the technical feasibility of the flagship North Lanarkshire AI growth zone, and that its renewable energy commitments had no realistic prospect of being met. At Blyth, a promised 4,000 jobs was quietly revised down to around 400.
These are not manufactured objections, and they are not nimbyism. When your own flagship site turns out to have had renewable commitments nobody ever expected to honour, you haven't been ambushed by objectors. You've handed them the argument, wrapped, with a bow on it. Households and farmers are being asked to cut water use while an industry that runs on enormous quantities of it gets waved through planning on numbers that don't survive contact with a journalist. I'd be furious too.
But here's the thing I keep coming back to. The upside being talked about isn't small money for a small risk. A Google-commissioned study estimates AI could add more than £400 billion to the UK economy by 2030, the equivalent of an extra 2.6% of annual growth, enough on its own to end the stagnation we've been stuck in for years. PwC's estimate has England's GDP up to 10.6% higher by the end of the decade. The government's own figures say the global AI chip market will be worth a trillion dollars in the early 2030s, and that 5% of it would bring $50 billion into Britain. Set against those numbers, £3.5 billion doesn't look ambitious. It looks like we've priced in losing before we've tried to win.
There's a harder problem underneath all this, and it's the one I'd put in front of Burnham first.
By the government's own admission, some data centre projects are waiting fifteen years for a grid connection. British industrial electricity costs around three times American levels, among the most expensive in Europe. And 80% of the UK's operational data centre capacity sits in London.
You cannot fix a fifteen-year queue with a taskforce. You cannot fix it with a Cabinet-attending minister either, however good he is, and Narayan does seem to be good. Compute follows power, and power follows planning and grid connections, and those are slow, unglamorous, deeply local fights that a reorganisation at the centre of government does not touch.
Which is precisely why this should be Burnham's problem to enjoy rather than his problem to dread. His entire political identity is about pulling power out of London and pushing it into the rest of the country. Eighty per cent of the nation's compute is in one city. If "Manchesterism" means anything at all, it means that number. The AI Growth Zone electricity discounts already skew north, £24 a megawatt hour in Scotland, £16 in Cumbria, £14 in the North East, from April 2027. That's the right instinct. It's just far too small, and it arrives too late, and it sits alongside a planning system that took his own flagship Scottish site and turned it into a scandal.
There's a version of this argument that's been made before, at a different hinge point, and it's worth being precise about rather than gesturing at.
The Industrial Revolution didn't make everyone richer overnight. Economic historians estimate British income per person barely moved from about $400 in 1760 to $430 in 1800, and didn't really take off until 1830 to 1860, when it jumped to $800. Real wages lagged productivity for decades, a gap economists still call Engels' Pause. The people objecting to mills and factories in 1780 weren't wrong that the changes were disruptive and unevenly felt. They were wrong that resisting it would stop it happening anyway. And the countries that industrialised late but seriously, Germany, Japan, eventually the US itself, ended up overtaking Britain, the country that got there first and grew complacent.
The economist Alexander Gerschenkron studied exactly this pattern. Latecomers to an industrial transition can leapfrog the countries that went first, he argued, but only if they mobilise resources and state backing at a scale the pioneers never needed, precisely because they're playing catch-up.
Britain doesn't get to be the pioneer in AI. That ship sailed while we were writing action plans about it. What we can still be is a serious fast follower. But fast followers don't get there by being cautious, and they don't get there on £3.5 billion.
Two weeks in, he has done the easy half properly and the hard half not at all.
The easy half is structure, and to be fair it's the half most governments never manage. AI at Cabinet. A taskforce reporting to the PM. Vallance running it. The Security Institute brought into the centre. That's a genuinely better machine than the one he inherited, and it happened at a speed that suggests somebody in that building understands the stakes.
The hard half is money and electricity and planning permission, and on all three the position is exactly what it was on the 19th of July.
So here's the test, and it's a simple one. The ten-year plan is due later this year. Burnham has told us it will lay out a path from where we are to where he thinks the country should be. There is one question worth asking of it when it lands.
Does it contain a compute number?
Not a taskforce. Not a minister. Not a strategy, a framework, a review or a growth zone. A number, with a date and a budget line next to it, that says this is how much sovereign compute Britain intends to own by such-and-such a year, and here is what we are spending to get it. Everything else is org chart.
Every month that number doesn't appear is a month closer to that child growing up into whatever we built, or failed to build, in the meantime. Not as an abstraction. As their actual job market, their actual wages, their actual chance of working somewhere that owns what it builds rather than renting someone else's platform on someone else's terms.
Mr Burnham, you asked for the job of rebalancing power in this country. You said so yourself, on the steps of Downing Street. You've since given AI a seat at your Cabinet table, which is more than your predecessor managed, and you deserve to be told that plainly.
Now put a number next to it.
Are we going to build the economy the next generation grows up inside? Or are we going to keep renting one from somewhere else, and hope nobody notices the difference until it's too late to change it?
]]>