Britain’s AI Offshoring Moment: When Electricity Prices Become An Export Policy

One in five UK firms say they’ve already moved AI workloads abroad. Congratulations, your national AI strategy now has a unit price per kilowatt-hour.

A survey-backed report argues high UK energy costs are pushing AI workloads out of the country, even as the government tries to sell ‘sovereign AI’ as an economic growth plan. This is the infra beat in its purest form: policy can talk about ‘innovation’, but the grid talks in lead times and bills.

The Claim

The Register reports on a CUDO Compute “Land, Power, Compute” survey of senior AI decision makers across the US and Europe. The headline number for the UK is brutal: around one in five firms say they have already moved AI workloads abroad because of high energy costs, and a third say energy costs are limiting their ability to scale.

Yes, the report comes from a rent-a-GPU company with clear incentives. But the constraint it describes is not exotic. It’s the same pattern Horizon keeps tripping over: GPUs are the shiny artifact, but power is the gate.

Why This Is Happening (and why it’s not “cloud strategy”)

There’s a comforting myth that “cloud” means location doesn’t matter. In practice, AI workloads have a physical address. They sit in data centers, behind grid connections, attached to permits, and priced by whatever electricity market you were unlucky enough to build in.

The UK has famously high electricity prices, and Full Fact has covered the comparative picture. When you combine high marginal prices with slow grid connections and planning delays, you get a quiet form of capital flight. Not the dramatic kind with suitcases, the boring kind with Terraform scripts.

Sovereign AI Meets The Invoice

Governments love the phrase “sovereign AI” because it sounds like independence and GDP. Businesses love it until it arrives wearing a bill. The Register notes the UK government is trying to blunt the influence of volatile gas prices on electricity pricing via long-term fixed contracts and other measures. That is real policy. It is also, in practice, a bet on whether prices come down fast enough to matter for this deployment cycle.

Meanwhile, companies are doing what companies do: arbitrage. If Eastern Europe, India, or the US looks cheaper or faster to provision, workloads move. The sovereignty story becomes a procurement story. The procurement story becomes a buildout story. The buildout story becomes a transformer story. This is the new national strategy: who can get power to a rack on time.

The Hidden Second-Order Effect: Where The Talent Goes

When workloads move, the operations teams, vendors, and adjacent service ecosystems tend to move too. Not immediately, not always, but eventually. Data centers create gravity. If the UK ends up with “AI policy” but not “AI power,” it risks building a research narrative on top of a deployment reality that lives elsewhere.

That’s how you get the modern version of offshoring: not call centers, but clusters. The GDP impact won’t arrive as a cliff. It arrives as a thousand small “we provisioned this region because…” decisions that compound.

The Singularity Soup Take

The UK’s AI ambition is not doomed. But it is being priced. The winners of the next phase are the jurisdictions that treat power, permitting, and grid connections as first-class industrial policy, not as someone else’s problem. “Sovereign AI” without sovereign electrons is just branding.

What to Watch

  • Concrete price mechanisms: do long-term contracts actually lower effective costs for energy-hungry compute users, or mostly stabilize headlines?
  • Grid connection lead times: are UK interconnection queues shortening in a way data-center developers can bank on?
  • Regional divergence: which UK regions can realistically host new capacity (and what they demand in return).
  • “Sovereignty” procurement: do government buyers pay a premium for UK/regionally controlled compute, or does cost win quietly?