Forget model benchmarks for a second. The biggest AI deal in the news this week isn’t about chips or chatbots — it’s about electricity. Britain’s National Grid is investing $1.75 billion for a 35% stake in Joulent, a U.S. energy platform building power infrastructure specifically for AI data centers.

Their first flagship project? Kilby — a 2.67-gigawatt gas-fired plant in West Texas, tied directly to a Microsoft-operated data center. That’s not a typo. One data center campus, drawing power on the scale of a mid-sized city.

Why Is a British Utility Buying Into Texas Gas Plants?

Here’s the thing: the AI industry has quietly changed what it competes on. In 2023, the bottleneck was GPUs. In 2024-25, it was data and talent. In 2026, it’s kilowatts. The companies that can secure firm, round-the-clock power near fiber routes win; the ones that can’t, wait in interconnection queues for years.

Think of it like a gold rush — except this time, the people selling shovels are utilities. National Grid saw the demand curve and did the math: why fight for AI upside in crowded software markets when you can own the one input every AI company must buy?

The numbers behind this shift are staggering. A 2.67 GW plant like Kilby produces roughly the output of two-and-a-half nuclear reactors. And it’s being built for, essentially, one customer’s workloads. Multiply that across every hyperscaler — Microsoft, Google, Amazon, Meta — each racing to lock in multi-gigawatt capacity, and you understand why energy access has joined regulation and national security as part of the same global contest.

So yeah. The AI race is now an infrastructure race, and infrastructure moves at the speed of concrete, not code.

What This Means For You

So what does this mean for you? If you work in tech, expect the geography of jobs to shift. West Texas, parts of the Midwest, anywhere with cheap land, gas, and grid capacity — these are becoming AI boomtowns. Data center construction, power engineering, grid software: these are the picks and shovels of this cycle, and they’re hiring.

If you’re an investor (and to be clear, this is context, not investment advice), notice what the smart money is doing: buying the layer below AI. Utilities, transformers, turbines, cooling. In my experience, when a hype cycle matures, the durable returns migrate from the shiny top layer down to the boring infrastructure underneath. It happened with railroads, with the internet backbone in the 2000s, and it’s happening again.

If you just use AI tools day to day? Honestly, this surprised me too: your subscription price is increasingly a function of natural gas prices and grid interconnection queues. When compute gets cheaper, features get better and prices hold. When power gets scarce, the opposite.

What Happens Next

Not everyone is cheering. And honestly, they have a point. A UN scientific panel just warned that AI development is outpacing both scientific understanding and government oversight — flagging risks from agentic systems, deceptive model behavior, cyber misuse, and biological threats. Building gas plants to feed models we don’t fully understand is, to critics, exactly backwards. Environmental groups add another uncomfortable question: whatever happened to those corporate net-zero pledges? A 2.67 GW gas plant is a lot of carbon to reconcile with a sustainability report.

But wait — there’s a counter-current. The same demand that’s reviving gas is also funding the largest wave of nuclear, geothermal, and grid-storage investment in decades, because hyperscalers want clean firm power almost as badly as they want fast power. The next 18 months will tell us which trend wins.

Watch for three signals: more utility-hyperscaler joint ventures like National Grid–Joulent, U.S. permitting reform aimed at speeding grid connections, and whether Microsoft’s competitors respond with their own captive power deals. My bet? All three happen before the year is out.

Key Takeaways

  • National Grid is paying $1.75 billion for 35% of Joulent, a U.S. platform building power infrastructure for AI data centers.
  • The first project, Kilby, is a 2.67-gigawatt gas plant in West Texas serving a Microsoft-operated data center.
  • The AI bottleneck has moved from GPUs to electricity — energy access is now competitive strategy.
  • A UN panel warns AI development is outpacing oversight, adding regulatory risk to the buildout.
  • The investment opportunity — and the jobs — are shifting to the infrastructure layer beneath AI.

Here’s my question to you: would you rather own the AI model or the power plant that runs it? Tell me in the comments — I suspect the answers say a lot about where we are in this cycle.