Every week we here a new 'AI' is worth billions. Harvey for example is hailed as the best 'AI' for the legal industry. It is worth eleven billion dollars. It writes contracts, runs due diligence, handles compliance. More than a hundred thousand lawyers use it. And it does not own a single AI model.
It runs on OpenAI, Anthropic, Google, whoever is best that quarter. Legal expertise on top, careful engineering around it, teams sat inside the law firms making it actually work. That is the company. That is the eleven billion.
Sequoia's Pat Grady has now backed Harvey three times. He put the value plainly: it is in the craft, the taste, the judgment of how you apply a model to a real job. Not the model itself.
Innovation such as this, is becoming world-class and dominating a lucrative market - hold that thought.
A provincial ambition
Britain has decided it wants to build the model and the machine.
The manifesto is grand. AI maker, not AI taker. A Sovereign AI Fund. A £1.1 billion hardware plan. £2 billion to grow compute twentyfold by 2030. A national supercomputer. A homegrown frontier model, Lumen, due late this year.
Read the speeches and you would think we were about to take on the world. Read the balance sheet and it looks provincial.
Here is the other side of the table. OpenAI and Anthropic alone are worth more than a trillion dollars between them. The American firms bankrolling them are worth trillions apiece. America's five biggest cloud companies are spending more than seven hundred billion dollars on AI infrastructure this year. Stargate is a five hundred billion dollar project. China has a trillion-yuan plan.
Against that, the grand British ambition amounts to persuading a handful of chosen companies to come cap in hand for a fraction of a fraction. It is not a challenge to America and China. It is a tribute act.
Put the two side by side. Britain's entire AI commitment, every fund and supercomputer and compute pledge added together, comes to less than one percent of what America's five biggest cloud firms will spend on AI infrastructure this year. Around two and a half days of it.
Are we backing the wrong horse? We may lose before the first jump (aka the next round of funding)... This article contrary to its tone, is not set out to criticise, it's a way to ask for a better conversation and get people to perhaps see there are better ways to IP and UK growth through AI, or AI enabled companies. Just as Harvey alone being significantly more value than the funding of an entire country. There's a bigger reason to invest in the sectors, not the model...
A target that moves every week
Say we somehow won it. What exactly would we have won?
The frontier moves weekly. The best model today is not the best model next month. Whoever leads gets overtaken, then overtakes again, in a loop funded by budgets no government on earth outside two countries can match. Build your national identity on being best at that, and you have built it on sand that shifts by Friday.
And then there is the bill.
Hundreds of billions in spending has to be paid back by someone. That someone is every business and household that ends up renting the result. Look hard at whether those numbers ever meet in the middle. A lot of serious people suspect they do not, and that what we are watching is the largest bubble in living memory inflating in real time.
The bill has already started to arrive. The buildout is now so large that economists have tied it to the price of everyday goods, and central bankers are watching data centre spending as a live inflation risk. The cost is not waiting politely for the technology to mature. It is landing on ordinary people now, before most of them have seen a single benefit. Chasing into the centre of that is not bravery. It is standing closest to the thing most likely to pop.
Somewhere along the way, "we use AI well" quietly became "we built an AI."
The government's own Sovereign AI Fund says it out loud, on its own website. The UK does not need to build every model. Sovereignty, it says, is about leverage, not control. There is no UK tech stack without the US. So the people writing the cheques already know we cannot own the frontier. Good. That is the honest position.
A fair critic will push back here, and they are half right. The fund does not only chase models. Its first bets are companies like Isomorphic Labs and Ineffable Intelligence, drug discovery and new learning algorithms, application-layer work of exactly the kind worth backing. Founders keep their own IP. The government has opened itself as an early customer through a procurement programme, which is one of the smartest moves in the whole plan. Credit where it is due.
Spend the sovereignty budget on what sovereignty actually means: data governance, security, and the freedom to change supplier, not a supercomputer built to run a model that will be obsolete before it is warm.
So the problem is not that Britain ignores innovation. The problem is proportion, and narrative.
The centrepiece of the strategy is three quarters of a billion pounds for a national supercomputer. The flag on the whole thing reads "sovereign." The speeches promise a homegrown frontier model. The biggest, loudest, most expensive commitments cluster around building the model. The application bets, the part the market actually rewards, get a thinner slice and none of the billboard.
That is the misallocation. Not innovation ignored, but innovation buried under a sovereignty flag it never needed, while the headline money points at the losing fight. And what filters down from that story is worse.
The thin boast
Somewhere along the way, "we use AI well" quietly became "we built an AI."
You see it everywhere now. Small companies, good companies, feeling they have to wave a card that says we made our own model, we have proprietary AI, we are an AI company now. It is naive. It is a bit desperate. And it is not their fault.
They caught the manifesto. They were told the model is the prize, so they reach for it, because they need a point of difference and nobody mentioned the prize is a mirage.
Same masquerade as the national one, just smaller. Dressing up for a race that was settled by people with a thousand times the budget.
The £11 billion wrapper
Go back to Harvey.
The dismissive line is that Harvey is "just a GPT wrapper." Fine. Then be a wrapper. The wrapper is where the eleven billion is.
The model is a commodity. It gets cheaper and smarter every few months, and it belongs to someone else no matter what you spend. What does not commoditise is the judgment. Knowing which problem is worth solving. Knowing what good looks like. Building the thing that lands in someone's actual workflow and saves them a Tuesday.
The whole British AI programme is valued at less than half of Harvey, a single legal app built on other people's models.
That is not the lesser prize. That is the prize. It always was.
The win is the platform, not the tech of the week
This is the part I want SMEs, entrepreneurs, scientists and creatives to hear, because it is the only bit that pays wages.
Stop chasing the model. The model of the week will always belong to someone bigger. Build the platform instead. The system, the taste, the workflow, the data, the judgment that sits around the tech and turns it into something a customer actually pays for. That is yours. That compounds. That does not get obsoleted by a press release from California on a Tuesday morning.
The frontier labs are renting you a commodity and calling it the future. The future, for almost everyone reading this, is what you build on top of it.
The race for AI Soverignty isnt the model, or machine
For the country. Fund the next Harvey in law, health, property, agriculture and the creative industries, the verticals where Britain already has world-class expertise and genuinely proprietary data. Keep the two best ideas already in the plan and scale them hard: let founders keep their IP, and use government procurement as a first customer so British AI companies earn revenue, not just grants. Treat frontier compute the way you treat electricity, something you buy at the best price and switch when a better deal appears, not something you go broke trying to 'own'. Spend the sovereignty budget on what sovereignty actually means: data governance, security, and the freedom to change supplier, not a supercomputer built to run a model that will be obsolete before it is warm.
For a business. Pick the problem, not the press release. Build the platform around it, the workflow, the data, the judgment, the deployment, the part that is yours and compounds. Use whatever model is best this week as a commodity input, and stay ready to swap it out. Where privacy or economics justify it, run capable models on hardware you own. Measure success by the problem solved, not the technology name-dropped.
None of this needs a trillion dollars. Back the application layer, not the frontier. Fund the next Harvey in law, health, property, agriculture and the creative industries. Rent compute like electricity, own your data and your deployment, and measure success by the problem solved, not the tech name-dropped.
The win is the platform, not the tech of the week.
