The AI race is being framed as a talent competition.
Startups. Universities. Regulatory frameworks. Investment funds. Britain scores reasonably well on most of those dimensions, and the government's ambition to position the country as a global AI leader is not unreasonable on the face of it.
But there is an infrastructure layer beneath all of it that the public conversation is almost entirely ignoring. And that layer has a single, irreducible input: electricity.
The Hardware Reality
AI is not a software business in the conventional sense. The large language models reshaping professional work require extraordinary compute to train and to run. That compute lives in data centres. Data centres consume electricity — vast, continuous, dispatchable electricity — at a scale that makes them among the largest industrial energy consumers in any economy.
This is not a detail. It is the constraint around which every serious AI infrastructure decision is being made.
Microsoft's decision to restart Three Mile Island in Pennsylvania — a retired nuclear plant that had been offline for five years — was not a PR exercise. It was an infrastructure decision driven by one question: where can we source 24/7 firm power at a price that makes the economics of AI infrastructure viable at scale?
That question is being asked simultaneously by every hyperscaler and every sovereign AI programme globally. Amazon, Google, Meta, Microsoft are all signing long-term power purchase agreements, building dedicated generation capacity, and making multi-decade site decisions based primarily on the answer to that question.
The answers are not pointing towards Britain.
The Competitive Reality
UK industrial electricity prices are among the highest in the developed world. The gap with the United States is not a temporary market dislocation — it is structural, embedded in grid levies, policy costs, and the particular way Britain has chosen to finance its energy transition.
American hyperscalers building data centres in Texas, Virginia, or Georgia face electricity costs that are a fraction of what a comparable facility in the United Kingdom would pay. The differential is not marginal. It is the kind of gap that changes investment decisions.
The Gulf states are competing for the same hyperscaler capital with cost structures that are different by an order of magnitude. The UAE has made AI infrastructure a national priority and is backing that ambition with energy economics that Europe cannot match. Saudi Arabia is doing the same. These are not abstract competitors — Microsoft's partnership with G42 in Abu Dhabi, and the scale of sovereign AI investment across the Gulf, represent real capital that could have been allocated elsewhere.
China's AI infrastructure buildout benefits from subsidised industrial electricity, particularly in provinces where hydro or coal power is abundant and cheap. Whatever one thinks of China's broader industrial policy, the energy economics of its AI programme are not subject to the same structural cost pressures Britain imposes on itself.
The Sequencing Problem
The critique of Ed Miliband's energy policy is not that decarbonisation is wrong, or that the renewable buildout is misconceived. The critique is sequencing — and the failure to acknowledge a fundamental incompatibility between the chosen transition mechanism and the demands of AI infrastructure.
Offshore wind produces electricity when the wind blows. AI data centres require electricity continuously, at any time of day, in any weather, at whatever scale the workload demands. The gap between intermittent renewable generation and firm, dispatchable power has to be bridged by something — either storage, which remains expensive and immature at grid scale, or backup gas generation, which adds both cost and carbon.
The grid balancing costs of a system built primarily around intermittent sources fall, disproportionately, on large industrial users. The policy levies embedded in UK electricity bills — funding contracts for difference, the capacity market, network charges, the renewable obligation — compound into a structural burden on energy-intensive businesses that is not present to the same degree in the United States, the Gulf, or most of Asia.
The result is that Britain is asking AI infrastructure investors to pay a premium for the privilege of being located here. That premium must be justified by something — regulatory stability, talent access, market proximity, government co-investment. Whether those factors outweigh a structural energy cost disadvantage of this magnitude is at minimum an open question. The honest answer is that they almost certainly do not, at the scale required to compete seriously.
A credible AI industrial strategy would engage with this directly. It would begin by asking what the firm power requirements of an AI data centre look like at scale, and then build an energy policy capable of meeting them at a globally competitive cost. That conversation requires honesty about what intermittent renewable generation alone cannot deliver — and the political courage to say so plainly.
The Nuclear Question
Britain has an answer to the firm power problem. It chose to close it.
The decision to phase out nuclear power — a source of firm, 24/7, low-carbon electricity — was made across decades by governments of both parties. It was not a policy driven by economics. The consequence is a grid increasingly dependent on sources that cannot guarantee the continuous output AI infrastructure requires.
Hinkley Point C is under construction. It is also massively over budget, years behind schedule, and a demonstration of what happens when nuclear procurement is treated as a one-off engineering project rather than a standing national capability. Small Modular Reactors are at the policy discussion stage — which in British infrastructure terms means they will not be generating power before 2035 at the earliest.
France, by contrast, has some of the cheapest industrial electricity in Europe. Firm, low-carbon, available around the clock. That was not an accident. It was the deliberate consequence of building and maintaining a nuclear fleet over decades — a decision that now gives France a structural competitive advantage in attracting exactly the kind of energy-intensive industrial investment that Britain is trying to court.
The United States is restarting retired nuclear capacity, permitting new SMR projects, and accelerating timelines across the board. The speed reflects a government that has made a decision: AI infrastructure is a national strategic priority, and energy policy will be priced accordingly.
What It Would Actually Take
None of this is irreversible. Energy policy decisions that take a decade to show in infrastructure prices take a decade to fix. But the fix has to start with an accurate diagnosis.
The diagnosis is not that Britain lacks AI talent, regulatory ambition, or investment appetite. The diagnosis is that the physical, economic foundation beneath every AI ambition — the energy infrastructure layer — is not competitive at the margin that matters for capital allocation decisions.
Addressing that requires a nuclear programme that moves faster than planning cycles permit. It requires a levy structure that does not place the entire cost of the energy transition on the industrial users Britain most needs to attract. It requires an honest public acknowledgement that electricity generation targets and firm, dispatchable power are not the same thing — and that the difference is measured in data centres that get built somewhere else.
Every AI strategy is ultimately an energy strategy.
Britain has not yet written the energy strategy that its AI ambitions require.