America is building AI infrastructure at city-scale while Britain's AI minister admits his country can't even plug in what they've already planned.
The Summary
- The US has dozens of massive AI data center projects underway, including a Texas campus pursuing 17 gigawatts and an Ohio facility targeting 10 gigawatts mostly contracted to OpenAI — each requiring more power than most American cities use
- UK data centers face 10+ year grid connection waitlists, forcing at least one planned facility to propose running on gas generators while waiting for electricity infrastructure to catch up
- Britain's AI minister publicly acknowledged the country lacks "sovereign level" AI processing capacity and must "move faster" on construction
- The divergence reveals which nations are prepared to become AI superpowers and which are stuck arguing about permits while the infrastructure gap widens
The Signal
The scale of American data center buildout has crossed into territory that rewrites regional economics. A single gigawatt of continuous electricity could power roughly 835,000 average American homes. The largest US projects now dwarf that benchmark. We're talking about facilities that consume more power than mid-sized American cities, built specifically to train and run AI models.
The US already operates more than 3,000 data centers with 1,500+ more under development, concentrated in rural areas where land is cheap and power infrastructure can be built without demolishing existing development. These aren't upgrades to existing facilities. These are greenfield projects, city-sized campuses rising in places that were farmland two years ago.
"US data centers accounted for nearly 40% of global data-center electricity use in 2025."
Meanwhile, Britain is discovering that ambition without infrastructure is just a press release. One planned UK data center has proposed running primarily on gas generators while waiting a decade or more for a firm grid connection. Not as backup power. As primary power. Because the electricity grid can't deliver what the facility needs, and won't be able to for ten years minimum.
This isn't a temporary bottleneck. It's a structural failure. AI minister Kanishka Narayan admitted the UK needs to reach "sovereign level" AI processing capacity, implying the country currently lacks it. When a government official says a nation must "move faster," what he means is: we're losing, and we know it.
Key contrasts:
- US: Building 10+ gigawatt facilities that already have power commitments
- UK: Can't connect planned facilities to the grid for a decade
- US: 1,500+ data centers in development pipeline
- UK: Debating whether to let one run on gas generators
The gap isn't closing. It's accelerating. Every month the UK spends negotiating grid connections is a month American facilities are training models, running inference, and locking in competitive advantages. Infrastructure determines destiny in the agent economy. You can't run frontier AI on goodwill and policy papers.
This matters because data center capacity is the physical substrate of AI capability. OpenAI didn't contract for most of a 10-gigawatt Ohio facility because they like the weather. They did it because training runs for the next generation of models require that scale, and if you don't have it, you don't compete.
The Implication
If you're building AI companies, look at where the power is going, not where the press releases are coming from. The US advantage in AI isn't just talent or capital. It's the willingness to build infrastructure at the scale the technology actually requires, in places where local governments say yes instead of forming committees.
For policymakers outside the US who talk about AI sovereignty: you're about ten years and several gigawatts behind. The time to build this infrastructure was five years ago. The second-best time is right now, but it requires approving projects that make people uncomfortable and writing checks that make budget offices nervous. Britain's AI minister knows this. The question is whether anyone with approval authority is listening.