AI’s Power Problem Is a Uranium Trade

The AI trade most investors own stops at the chips. The trade fewer are watching starts one link earlier, at the power plant, because the binding constraint on artificial intelligence is not compute but the electricity to run it, and the cleanest continuous source of that electricity increasingly points to nuclear. As of September 2026, several experts interviewed on Wealthion argue that AI’s power problem is, followed to its root, a uranium and nuclear thesis. Here is the chain of reasoning, and where they say it leads.

Why is electricity the real constraint on AI?

Because a data center is a machine for converting power into computation, and the power bills are staggering. Brett Rentmeester of WindRock Wealth traced the scale in his July interview, pointing out that the honest way to size the AI boom is not by counting subscriptions but by counting the physical buildout: “back in 2023, if we base spending not on people doing AI subscriptions, but instead on data center spend” you see the true commitment, and that spend has since climbed into the hundreds of billions. He cited the backlogs directly, including “a $460 billion backlog of orders” at one hyperscaler’s cloud unit. What all that capital buys, in his words, is blunt: “They need data center and compute,” and behind compute sits the power to feed it. The power draw is the part that does not shrink: a chatbot query can be made more efficient, but the aggregate demand curve for data-center electricity has bent sharply upward, and utilities are the ones who have to answer it.

Why does AI power demand point to nuclear?

Because data centers need baseload power, electricity that runs continuously, not just when the wind blows or the sun shines, and nuclear is one of the few carbon-free sources that delivers it. Steven Feldman made the physical point on Wealthion in his June interview: the investable edge of the AI story is in the things you plug the machines into, “it’s the uranium that’s has the small nuclear reactors that people need next to these things” [as spoken]. Small modular reactors, purpose-built to sit beside a data-center campus, are the clearest expression of that idea. As of September 2026, the interest in them is why several technology firms have begun signing nuclear supply and reactor-restart deals directly, contracting for power years before the reactors exist, a striking sign of how seriously the buyers take the coming shortfall. Feldman’s second point is geopolitical: governments will increasingly want the fuel onshore, because “they better have a local supply” of whatever powers the economy.

Is this just an energy story, or a supply story?

Both, and the supply half is what makes it a trade rather than a theme. The resource investors Nick Lawson and Ben Finegold built their entire “Molecular War” thesis on Wealthion, in their May interview, around the fragility of strategic-material supply chains, and they put uranium at the center: “the core thesis for us at the moment is is based around predominantly uranium.” Their reasoning connects the demand and supply sides in one image. A nuclear plant cannot flex its fuel: “If you don’t have uranium in it, it shutters” and costs enormous sums to restart, which makes utilities structurally desperate to secure supply. Yet supply cannot answer quickly, because “you cannot have a supply response within 2 years at least” from new mines. Rising, inflexible demand meeting slow, concentrated supply is the definition of a squeeze.

How big is the demand actually?

Large enough that the existing fleet is already stretched before the AI surge fully lands. The United States runs on the order of 100 gigawatts of nuclear capacity today, and the AI buildout is arriving on top of a grid that was not planned for it. Mark Mills, in his July interview, has made the broader case that the physical inputs to computing, power above all, have been chronically underbuilt, the same underinvestment argument covered in our analysis of what AI actually needs. The demand is not speculative; it is contracted backlog and announced capacity, arriving faster than new generation can be built. That timing mismatch, demand landing now against generation that takes years to build, is what turns an energy story into a scarcity story with an investable edge.

Where does this leave the value chain?

At the same place our uranium supply-deficit analysis lands, viewed from the demand side. The AI power thesis does not resolve into a single asset; it resolves into a chain. There is the uranium itself, whose supply cannot respond for years. There are the fuel-cycle chokepoints, conversion and enrichment, where capacity is scarce and geopolitically concentrated. There are the reactor and small-modular-reactor builders that turn fuel into power. And there are the utilities and independent power producers that sell the electricity into the data-center boom. Each sits at a different point on the chain, with a different exposure to the AI demand shock. The investors above are not buying AI power as a slogan; they are identifying which link in the chain is most constrained and hardest to replace.

What could weaken the thesis?

The honest risks are demand and efficiency. If AI power needs prove smaller than the current backlogs imply, or if the industry’s efficiency gains, better chips and cooling, reduce power draw per unit of compute, the downstream pull on nuclear and uranium softens. Rentmeester himself frames AI’s economics as an open question, noting the enormous capital going in against uncertain returns. And the fastest grid response to a power crunch is often the cheapest available fuel, not the cleanest, which is why the same investors note gas and even coal can answer first. The nuclear-and-uranium leg is the durable answer, not necessarily the immediate one. An investor persuaded by the thesis still has to decide whether to own the fuel, the fuel-cycle chokepoints, or the power producers, because the AI power shock hits each of those links differently and on a different timeline.

FAQ: AI, Power, and Uranium in Brief

Why does AI need so much electricity? AI runs in data centers that convert power into computation at massive scale. The buildout is measured in hundreds of billions of dollars of data-center spending and multi-hundred-billion-dollar cloud backlogs, all of which require continuous power.

Why is nuclear power linked to AI? Data centers need baseload power that runs continuously. Nuclear is one of the few carbon-free sources that provides it, so the AI buildout has driven new interest in reactors, including small modular reactors sited near data centers.

How does AI power demand affect uranium? Nuclear reactors run on uranium, and uranium supply cannot expand quickly. Rising, inflexible reactor demand meeting slow supply is why experts on Wealthion frame AI’s power needs as a uranium thesis.

What are small modular reactors? Smaller, factory-built nuclear reactors designed to be deployed faster and closer to demand, such as beside a data-center campus. They are a key reason AI power demand is being connected to nuclear.

What could go wrong with the AI-uranium thesis? Weaker-than-expected AI power demand, efficiency gains that cut power draw, or faster grid responses from cheaper fuels like natural gas could all soften the pull on nuclear and uranium.

Which experts and interviews does this article reference? Wealthion interviews from May to July 2026: Brett Rentmeester on AI’s $920 billion gamble, Nick Lawson and Ben Finegold on the Molecular War thesis, Steven Feldman on the AI resource rush, and Mark Mills on AI, oil, and underinvestment.

Wealthion editorial content is for informational purposes only and is not investment advice, and nothing here recommends any security. The views quoted belong to the named guests. If you want a professional read on how the AI infrastructure theme fits your own portfolio, you can request a free portfolio review at https://www.wealthion.com/advisors/.