August 12, 2026 · Mochatrade, Parth Maheshwari · 3 min read
Nuclear Is Becoming the Next Great US Market Trade. AI Left It No Choice.
AI cannot grow without power, and that is reviving an entire American sector. Restarted reactors, uranium, and small modular startups each sit at a different point on the curve.
In 1979, Three Mile Island suffered the worst nuclear accident in American history and became shorthand for everything people feared about the technology. In 2027, it is scheduled to switch back on. The customer is Microsoft. The reason is artificial intelligence. That one reversal captures a shift now moving through US markets.
AI does not only need chips. It needs power, far more of it than most people realise.
Why AI broke the power math
A modern AI data centre runs every hour of every day, and training the largest models pulls electricity on the scale of a small city. Global data centre demand is expected to climb from about 415 terawatt hours in 2024 to roughly 945 by 2030, more than double.
Wind and solar cannot carry that load alone, because AI cannot pause when the wind drops or the sun sets. It needs power that runs constantly. Nuclear does, more than 92% of the time, against roughly a third for wind and a quarter for solar.
So the companies with the deepest pockets went hunting for reactors.
The biggest private nuclear buying wave since the 1970s
By mid 2026, every major US tech giant had signed at least one nuclear deal. Together they have committed close to 9.8 gigawatts across thirteen projects. Microsoft locked a 20 year deal worth about $16 billion, roughly ₹1.53 lakh crore, to restart Three Mile Island. Amazon is converting a nuclear site into an AI campus. Meta has committed the most capacity of all, though its power arrives years later.
The clearest way to know a thesis is real is to watch who is writing the big cheques. Right now, every one of them is going nuclear.
How the sector could split
This is where it matters for an investor. The trade is not one thing, it is three, each on a different timeline.
The excitement has gone to small modular reactor names like Oklo and NuScale, the factory built machines meant to sit beside a data centre. They may be the future, but the first ones for AI are unlikely before the early 2030s.
AI needs power now. So the near term value may sit instead with the companies that already run reactors, names like Constellation and Vistra being switched on and pushed for every megawatt, and with uranium miners like Cameco, since every reactor must burn fuel.
The dream is the small reactor. The trade, for now, could be the old plant getting a second life.
What could break this
Restarts can slip, reactor costs may stay high, and if AI spending cools, long dated deals could be reopened. Uranium, near $84 a pound, can swing hard either way.
For years the AI race was about who had the best chips. The next stage may be settled by something older and far less exciting, who can plug in enough power to keep the machines running. That contest is playing out in US markets right now.
Educational content. Not investment advice. Figures approximate and as of mid 2026. Do your own research.
