To compete with the US on AI, the British need to cut energy bills

Share this article

Last Friday, the most capable AI model in the world went dark. Anthropic had released Fable 5 three days earlier; on June 12, a U.S. executive order cut off access for foreign nationals. With no clean way to wall Americans off from the rest of us inside a global system used by hundreds of millions of people, the model was switched off for everyone. No prior consultation of allies or stakeholders, no warning, just U.S. government fiat.

For Europeans and Britons, it should have a wake-up call – yet the news barely registered amid stories from Iran, G7, and the south lawn of the White House. Never mind the flimsy national security rationale, the story is a reminder that the frontier of the most consequential technology of our era sits far outside of European or British control – and is governed not so much by market forces but rather at the discretion of the U.S. (or Chinese) government.

The heavy-handed intervention by the Trump administration may be itself a part of a power play, seeking to assert government control (or perhaps partial ownership) over Anthropic, not unlike its earlier ‘deal’ with Nvidia and Intel.

In some ways, Europe’s best hope is that AI proves to be a flop – that the promised improvements in productivity, the new diagnostic methods and drugs, AI-powered logistics, or autonomous robots in manufacturing never materialize. It is still conceivable that a few years from now, investors who have poured trillions into advanced semiconductors, data centres, and software will only have a handful of hallucinating chatbots to show for their efforts. But that outcome is growing less likely by the day.

If, on the other hand, AI is akin to a general-purpose technology, Europeans and the British have a problem. The gap separating us from the United States is eyewatering. Fourteen of the world’s fifteen largest AI companies, starting with Nvidia (valued at $5 trillion) are listed in the United States.

Europe and the UK have no pure-play to set against them. Our two champions are ASML, the Dutch firm that monopolizes the lithography machines without which no advanced chip gets made – now also Mistral’s largest shareholder – and Mistral itself, France’s lonely frontier lab, valued in the low tens of billions against American rivals an order of magnitude larger.

The compute gap is just as stark. The United States accounts for around 44% of global data-centre capacity; the EU holds barely a fifth of that. Americans have more capacity under construction than Europe’s entire existing stock.

The four largest American hyperscalers plan to spend over $370 billion on data centres in a single year—more than the whole of Europe’s projected buildout for the coming decade. If AI turns out to be the general-purpose technology its boosters promise, a continent this far behind on both models and compute has a serious, compounding problem.

We are not going to conjure a European or a British lab to rival the giants at the frontier; that ship has, for all practical purposes, sailed. But it is not too late to protect ourselves against the worst-case scenario, however.

In a fictionalised form, the Europe2031.ai website outlines what the latter may look like: AI takes off as a massive driver of productivity growth and, at one point in the next five years, computing capacity (concentrated in the United States) becomes the binding constraint on its diffusion. With little capacity of its own, Europe is shut off from the most advanced capabilities and its companies rendered uncompetitive – the continent thus becomes a permanent economic backwater, at the mercy of Washington and Beijing.

To avoid that outcome, Europe and the UK must do two things. First, instead of trying to keep US AI away, they should become the best customers for US firms – both to ensure that absorbs whatever economic benefits the technology generates and also to give the continent economic leverage over a possible anti-European administration.

Second, we ought to become a reliable supplier of critical inputs that frontier AI cannot do without—above all, energy and compute. ASML is nice but a Europe that hosts a meaningful share of the world’s data centres would have even more to trade than just control over a chokepoint.

Here is the catch, and it is one we have inflicted on ourselves. The single biggest obstacle is the price of electricity. American and Chinese industry pays around 8 U.S. cents per kilowatt-hour. French industry pays roughly 12; German industry close to 18; and British somewhere between the two. You cannot run the most energy-hungry industry of the century at twice your competitors’ power costs and expect to win. Part of the gap is carbon pricing, part is our dependence on imported LNG – but most of it is driven by policy choices.

The call to action writes itself: more power, and fast. Fast-track every project that adds supply to the grid, and upgrade the grids themselves. Put the Nordic countries’ free cooling and abundant hydropower to work. And stop pretending that nuclear is ‘too slow and too expensive’ to bother with – that is not a law of physics, it is a description of our own permitting regimes. The cost of a new reactor is overwhelmingly a function of regulation, and regulation is the one thing our elected representatives are entirely free to change.

A Europe (and a Britain) that continues to retreat into digital autarky and regulatory pique will a poorer, more dependent version of what it already is. It is not too late to avoid that outcome – but the window to do something about it is closing fast.

Share this article