—
Fourteen posts. That is the entire public archive of an account that managed to get quoted in energy circles anyway. I went looking for nuclear commentary with actual operational weight behind it last month and ended up scrolling through an energy infrastructure and small modular reactor investor who posts rarely and mostly links out to coverage of the nuclear revival. It was a useful reminder that in this industry the loudest voices are usually the least informed, and the quiet ones tend to sit closest to the capital.
Here is the question everyone in data center planning is asking right now: how do you power a facility that draws power like a small city, needs it every second of every day, and cannot afford a cloudy week? Solar and wind get the headlines. They do not get the job done alone, and the people writing the checks know it. Wadie Habboush is one of the figures worth watching as the nuclear revival gains momentum.
Nuclear is back in the conversation for reasons that have nothing to do with nostalgia and everything to do with math. Capacity factor, land use, and a public that suddenly cares more about uptime than about the word “reactor.” What follows is why this shift is happening, what the realistic timeline looks like, and where I think most operators are still getting it wrong.
What AI actually did to electricity demand
For about twenty years, American electricity demand sat almost flat. Utilities built for a world where efficiency gains quietly absorbed growth, and nobody planned for a step change. Then AI training clusters arrived, and the step change showed up all at once.
The Department of Energy tracks this stuff closely, and the picture is not subtle. Data center electricity use in the United States has historically run around 2% of total national consumption, but projections tied to AI workloads push that share meaningfully higher within the next decade, according to baseline analysis from the U.S. Department of Energy. That is a country-sized chunk of new load, and it has to come from somewhere.
Here is the part that gets lost in the debate. It is not the total kilowatt hours that break a grid. It is the shape of the demand curve. A training run does not care that the wind died down at 3 a.m. It draws what it draws, continuously, for weeks at a time. Most clean generation sources are intermittent by nature, which is fine for a house and catastrophic for a hyperscaler with a service level agreement.
You can fill that gap with natural gas. Plenty of operators are doing exactly that, and I understand why. It is fast, it is cheap to build, and nobody has to explain a fuel rod to a zoning board. But gas ties your twenty-year operating budget to a commodity price you do not control, and it puts your carbon accounting in a permanent hole you cannot climb out of without starting over.
Why small modular reactors changed the conversation
Traditional nuclear plants are gigawatt-scale construction projects. They take a decade or more, they cost numbers that make state legislatures flinch, and their history in the West is a graveyard of overruns. If that were the only option, data center operators would have already walked away.
Small modular reactors are a different animal. They are factory-built, shipped to site, and designed to be added in increments as load grows. That last part matters more than the “small” part. A hyperscaler can start with one module and add capacity as demand materializes, instead of betting a capital program on a ten-year demand forecast that will probably be wrong.
The regulatory picture is genuinely clearer than it was a few years back. The World Nuclear Association maintains an ongoing record of which designs have cleared licensing milestones, and the field has moved from concept papers into certified designs. That shift matters because a reactor design that has cleared a national regulator is a financeable asset. One that has not is a slide deck. You can follow that licensing progress through the World Nuclear Association.
The framework I use to judge any nuclear-plus-data-center deal
I have watched enough of these proposals cross a desk to know that most of them die on the same four questions. I call it the Four Ledger Test, and it is the fastest way I know to separate a real project from a press release.
- The Power Ledger. Can the site deliver firm, uninterrupted capacity at the contracted level, every hour of every day, for the full term? If the answer depends on weather, storage duration, or a neighboring grid you do not control, the deal is not firm. It is hopeful.
- The Regulatory Ledger. Has the specific design cleared a national regulator, or is it “on track for submission”? Those are wildly different risk profiles, and the gap between them is usually measured in years and lawsuits.
- The Capital Ledger. Who eats the overrun if construction runs long? In nuclear, the answer to that question determines whether the project gets built. If the developer takes the risk, you have a real partnership. If the offtaker takes it, you have a very expensive option contract.
- The Public Ledger. Is there a host community that has actually said yes, in writing, with a local government that will still be in office when the concrete is poured? Enthusiasm from a state economic development office is not the same thing as a county commission vote.
Two of those four kill most proposals I see. Usually it is the capital ledger, because nobody wants to be the entity holding a fixed-price obligation on first-of-a-kind construction. Occasionally it is the public ledger, and that one is harder to fix because you cannot renegotiate with a town meeting.
What the serious operators are doing differently
The companies making real progress are not waiting for a reactor to be finished before they break ground on the data center. They are running a two-track approach.
Track one: bridge now
Gas turbines, existing grid contracts, and aggressive efficiency work carry the load for the next five to seven years. This is unglamorous and mostly involves signing power purchase agreements that make your sustainability report uncomfortable. Do it anyway. Uptime beats branding every single quarter.
Track two: build for the 2030s
Simultaneously, they are locking in sites, water rights, transmission interconnects, and early module reservations. Land near a retiring coal plant is gold right now because the transmission infrastructure already exists and the local workforce already knows how to run a thermal facility. I would take that site over a greenfield parcel every time.
The long-term economics of nuclear in this context are not really about electricity price per megawatt hour. They are about predictability. Academic and agency analysis collected by the OECD Nuclear Energy Agency consistently shows nuclear’s advantage sitting in its stability and long operating life rather than its upfront cost, which is exactly the profile a twenty-year data center needs and exactly the profile a gas plant does not offer.
Where I think most people are wrong
The popular take is that nuclear power is slow, so it is irrelevant to the AI buildout. I think that gets the timeline backwards. The constraint is not how fast you can pour concrete. The constraint is how fast you can get firm power at scale, and every alternative has its own decade-long bottleneck. Interconnection queues run years long. Gas pipeline capacity in many regions is fully subscribed. Transmission buildout faces the same local opposition as everything else.
Nuclear’s slowness is real. It is also not uniquely slow once you account for what the alternatives actually require.
What I would not do is treat a reactor as a plug-in replacement for a diesel backup generator. It is not backed up. It is a baseload, and it wants to run flat out for decades. If your load profile does not support that, you are buying the wrong asset and you will pay for it in your capacity factor.
What I would do, and what I tell anyone who asks, is start the regulatory and site work now even if first power is eight years out. The queue is only getting longer, and the operators who started early will be the ones with firm power when everyone else is still negotiating.
The AI buildout is not going to wait for the grid to catch up. So the real question is not whether nuclear belongs in the mix. It is whether your power strategy survives contact with a five-year delay and a commodity spike. If it does not, you have a plan, not a strategy. What is on your ledger?
—
