The Nuclear Race to Power AI

Zach Martin

Artificial intelligence may be the most important technology race of our generation. But there is something AI cannot manufacture. Electricity. Every model, every GPU, every hyperscale data center, every autonomous system, every robot and cloud platform needs enormous amounts of electricity, twenty four hours a day, seven days a week, whether the sun is shining or the wind is blowing. After decades of closures, political opposition, cost overruns and plain skepticism, nuclear is back at the center of the conversation. Amazon, Google, Meta, Microsoft, utilities, uranium producers, reactor developers, governments, and billions of dollars of capital are all pulling in the same direction. The old question was, should we build nuclear? The new question is, can we build enough, fast enough, to power the next industrial revolution? I'm Zach Martin, and this is NUCLEAR CAPITAL, the series premiere. The Nuclear Renaissance: Powering the AI Economy. A KMKT I R Hub Radio exclusive. And I'm joined by someone who will make sure I earn every claim I make, Emily Martin.

Emily Martin

Thank you, Zach. And I will take that job seriously. Here's how I see my role in this series. You bring the big picture and the energy. I'm going to ask the investor questions, push on the risks, and translate the jargon so nobody has to nod along to a term they don't understand. If I hear a megawatt, a gigawatt, or an acronym, I'm going to stop you. Before we go any further, one important note. This show is informational. Nothing we say is a recommendation to buy or sell any security. We talk about companies because they're part of a story, not because we're telling you what to do with your money.

Zach Martin

That's exactly right, and it's the right frame for the whole hour. I'd put it this way. Investing in a theme is like training for a marathon. The theme being real doesn't mean every runner finishes, and it doesn't mean the pace is easy. So we're going to walk the whole course today. The power problem, the Big Tech deals, why nuclear and why now, the fuel chain, small modular reactors, a watchlist of names, the bull case, the bear case, a scorecard, and then what to watch in the short term versus the long term. And at the end, I want to land on what I'm calling the second AI trade.

Emily Martin

Then let's start where the whole thing starts. Plain language. What is the power problem? Because I think a lot of listeners picture AI as software, something that lives in the cloud, and the cloud sounds weightless.

Zach Martin

The cloud is a factory, Emily. That's the thing to hold onto. The AI revolution is computational, but it is also physical. Data centers, transmission lines, transformers, cooling systems, chip fabs, gas plants, renewables, and increasingly nuclear reactors. The International Energy Agency has described electricity demand as entering a new age, with data centers and AI among the forces pushing consumption higher. And the numbers in the research we pulled are striking. Global electricity use by data centers is projected to grow from about 460 terawatt hours in 2024 to over 1,000 terawatt hours by 2030, and about 1,300 terawatt hours by 2035.

Emily Martin

Okay, terawatt hours. Translate that for me. Is that a lot?

Zach Martin

A terawatt hour is a billion kilowatt hours. So think of your home electric bill, then multiply by a number so large it stops meaning anything. The better way to read it is the direction. That 460 figure nearly triples by 2035. And here's a second number from the same research, which I'll attribute to Goldman Sachs. They estimate that meeting all the data center power growth expected by 2030 would take something like 85 to 90 gigawatts of new nuclear capacity. And they say well under ten percent of that will be available globally by 2030.

Emily Martin

Hold on. That's the first thing that jumps out at me, and it's a problem for the bullish story. If the need is eighty five to ninety gigawatts and less than ten percent shows up by 2030, then nuclear isn't solving the AI power crunch this decade. Something else is. What is it?

Zach Martin

Something else is, and I'll concede that right away. The same research says natural gas supplies over forty percent of electricity for US data centers, renewables about twenty four percent, coal around fifteen percent, and nuclear roughly fifteen to twenty percent. So gas is carrying a huge share, and it will keep doing that through the decade. Nobody serious says nuclear flips that overnight. My argument is about what the buyers want to be standing on in 2035 and 2045, and what they're willing to sign twenty year contracts for today.

Emily Martin

So the claim isn't nuclear wins the next three years. It's that the people with the biggest checkbooks are positioning for the long game.

Zach Martin

Yes. And the reason is a physics reason, not a political one. AI needs firm power. That's the term of art. Firm power means electricity that's there when you ask for it, not when the weather feels like it. A data center doesn't stop when the sun goes down. Training a model or serving millions of queries isn't something you pause for cloud cover. Nuclear can provide large quantities of dispatchable, low carbon electricity around the clock. The research puts nuclear capacity factors at roughly ninety two to ninety five percent or more, compared to about twenty five to thirty five percent for solar and wind.

Emily Martin

Capacity factor. Let me say it back and you tell me if I've got it. If a plant could theoretically produce a hundred units of power running flat out all year, the capacity factor is how many units it actually delivers over the year.

Zach Martin

Perfect. A nuclear plant is like the runner who shows up every single morning. Solar is brilliant at noon and absent at midnight. That's not a criticism of solar, it's just what it is. And for a data center, the midnight hours matter as much as noon.

Emily Martin

But let me push on that, because I can hear an objection. Solar plus batteries is getting cheaper. Gas plants can be built faster than any reactor. Why isn't the answer simply more of everything, quickly, and let nuclear be a footnote?

Zach Martin

Because the answer is all of the above, and I'd never argue otherwise. The grid will need everything. But there are two things nuclear does that the others struggle with together. One is density. The research says a multi reactor small modular facility takes roughly 50 acres, while equivalent solar or wind output can take thousands of acres plus the transmission to move it. The other is that the power is carbon free and firm at the same time. Gas is firm but it's not carbon free. Solar is carbon free but it's not firm. A company with a net zero pledge and a gigawatt campus is looking for both at once.

Emily Martin

And the cost? I've seen the Goldman numbers in what we pulled. Nuclear capital cost somewhere between about sixty four hundred and twelve thousand six hundred dollars per kilowatt, versus around twelve ninety for natural gas. That's five to ten times as expensive to build.

Zach Martin

It is, and that's the cost problem, and we'll come back to it in the bear case because it deserves real time. The research itself says the economics favor nuclear mostly where carbon free requirements justify the premium. What's changed is who is paying the premium and why. The buyer used to be a regulated utility worried about rate cases. Now the buyer is a company that makes tens of billions in profit and considers a delayed data center far more expensive than an expensive kilowatt hour.

Emily Martin

That's a good line, and I think it's the center of the whole thesis. Time to power is worth more than price of power. Let's get into the buyers. Walk me through the deals. And I want to know which are signed contracts and which are, as I keep seeing in headlines, up to something.

Zach Martin

Fair, and I'll mark that distinction as we go. Let's start with Google. Google partnered with Kairos Power on advanced reactors, up to 500 megawatts. The research describes it as the first corporate agreement to buy power from multiple small modular reactors, roughly six to seven reactors. The first project is targeted for 2030, with additional deployments through 2035. Under the structure, Kairos develops, builds and operates the plants and sells the energy to Google under power purchase agreements.

Emily Martin

Power purchase agreement. PPA. Explain it like I'm a retail investor who has never read one.

Zach Martin

A PPA is a long contract where a buyer agrees to purchase electricity from a plant at agreed terms for years. Think of it like a mortgage lender looking at a signed lease from a blue chip tenant. The lender can finance the building because the rent is predictable. A twenty year PPA from a company like Google or Amazon is the signed lease that lets a developer, a lender, or a utility say yes to a project that otherwise might be too risky to finance.

Emily Martin

I like that. And the Kairos piece has something concrete now, right? Not just a framework.

Zach Martin

Yes. In April 2026, Kairos broke ground on Hermes 2 in Oak Ridge, Tennessee. It's designed to supply up to 50 megawatts to the Tennessee Valley Authority grid, supporting Google data centers in Tennessee and Alabama. The research describes the Google data centers as sitting in Montgomery County, Tennessee, and Jackson County, Alabama. And it describes TVA as the first US utility to sign an offtake agreement with an advanced nuclear plant. The technology uses a molten fluoride salt for cooling and ceramic pebble type fuel.

Emily Martin

Fifty megawatts is small. Google runs a lot more than fifty megawatts of data centers.

Zach Martin

It's small on purpose. That's the first of a kind step. And the research makes a point I find clarifying about who carries the risk. Kairos and Google carry the financial risk of building the first project. TVA provides the revenue stream through the PPA. So the ratepayer is not footing the bill for the experiment. That's a new way to finance nuclear, and it's one of the things that makes this moment different.

Emily Martin

So the hyperscaler is basically the venture capitalist and the customer at the same time.

Zach Martin

That's a sharp way to say it. Now Amazon. Amazon has gone at this from several angles. First, small modular reactors with X-energy. Amazon has been developing a twelve reactor project in Washington State with Energy Northwest, targeted for the early 2030s. The research says it starts with four Xe 100 units, about 320 megawatts, and is designed to scale to twelve units and 960 megawatts. Amazon also took part in a roughly 500 million dollar Series C 1 financing for X-energy, alongside Citadel's Ken Griffin and others, as part of a goal of more than five gigawatts of X-energy projects by 2039.

Emily Martin

And what about the Amazon deals on the existing fleet? Because I know the one you've been itching to talk about.

Zach Martin

Calvert Cliffs. On September 30, 2026, Constellation and Amazon announced a twenty year power purchase agreement tied to Calvert Cliffs in Lusby, Maryland. This is the real contract, not an up to. The announcement says it supports more than three billion dollars of Maryland infrastructure investment, including improvements across the entire 1,790 megawatt plant, and about 190 megawatts of new, emissions free capacity from uprates coming online between 2030 and 2032. The agreement covers 690 megawatts of power, including that 190 megawatt uprate.

Emily Martin

Uprate. Another one I'd like decoded.

Zach Martin

An uprate is squeezing more output out of a plant that already exists, through equipment upgrades and engineering improvements, within the safety limits the regulator allows. It's the cheapest new megawatt you can find in nuclear, because you're not building a new plant. You're getting more from the one you've got.

Emily Martin

Now here's where I want to bring in a wrinkle, because it's in our research and it matters. Weeks earlier, in August 2026, Amazon abandoned its plan to build data centers right on the Calvert Cliffs campus. There was fierce voter backlash in the Calvert County primary in June. Three commissioners who had voted against a pause on data centers lost their races. So is this deal as clean as the press release makes it sound?

Zach Martin

It's a good catch, and it's a real lesson. The Maryland Matters reporting says Amazon dropped the proposal for multiple data center campuses at Calvert Cliffs, and Amazon said it was a business decision. The county planned a hearing on a temporary moratorium. And the September agreement is structured differently. The power from Calvert Cliffs keeps flowing to the PJM regional grid, the big multi state grid, just as it does now, and Amazon gets a related retail supply agreement to support its operations in that market. So the nuclear contract survived even though the data center on the plant's doorstep did not.

Emily Martin

Which tells me something I don't think the bulls talk about enough. The social license. You can have the most reliable reactor in the world and still lose a vote in a county commission primary. The same research says nearly seventy seven percent of Maryland's population lives under some kind of data center moratorium.

Zach Martin

That figure comes from the Chesapeake Climate Action Network via the Maryland Matters report, so I'd attribute it that way. But I agree with the substance. Local politics is a real variable. Where I land differently is on what it means for nuclear itself. The opposition there was aimed at hyperscale data centers, not at the reactor. The plant is still, in Constellation's words, Maryland's largest source of clean energy, roughly eighty percent of it, and the contract is meant to support relicensing it for another twenty years. I'd say the signal is, the data center siting fight and the nuclear value story are related but separable.

Emily Martin

Separable, but connected through the same buyer. If voters push data centers out of one county, the buyer moves to the next. Fine. I'll give you that for now, but I want it on the board as a risk. Let's keep going. Meta.

Zach Martin

Meta is the one that really tells you the customer base is changing. In January 2026, Meta announced agreements with Vistra, TerraPower and Oklo, building on an earlier deal with Constellation, that could support up to 6.6 gigawatts of new and existing nuclear capacity by 2035. Meta says that makes it one of the most significant corporate purchasers of nuclear energy in American history. Let me break it into three pieces. With Vistra, twenty year agreements to buy more than 2.1 gigawatts from the Perry and Davis-Besse plants in Ohio, plus uprates across Perry, Davis-Besse and Beaver Valley in Pennsylvania totaling 433 megawatts, expected in the early 2030s.

Emily Martin

So that piece is existing plants, and Meta is paying to make them bigger and keep them running.

Zach Martin

Right. Then Oklo. Meta is supporting a 1.2 gigawatt power campus in Pike County, Ohio. The research says it's planned as 16 Aurora Powerhouse units of 75 megawatts each, with pre construction and site work slated for 2026, a first phase of around 150 megawatts targeted as early as 2030, and the full 1.2 gigawatts by 2034. And Meta is providing prepayments for fuel. Then TerraPower, where Meta funds the development of two Natrium sodium fast reactors, 345 megawatts each, that can't start supplying power until at least 2032, with rights to up to six more units, 2.1 gigawatts, targeted by 2035.

Emily Martin

I want to flag something for the listener. Those words, as early as, targeted for, rights to, are doing a lot of work. Those are options on the future. The Vistra piece is much closer to bankable than the Oklo piece.

Zach Martin

Completely agree, and that's the right way to read the Meta announcement. I'd rank the certainty in layers. Existing plants and uprates are the most certain. Restarts come next. Then the first advanced reactors, then fleets. And the Microsoft deal belongs in that restart layer. Microsoft signed a twenty year agreement with Constellation to bring back Three Mile Island Unit 1, about 837 megawatts, renamed the Crane Clean Energy Center, targeted to reopen in 2028. The research puts the deal's value at about sixteen billion dollars. And to be clear, Unit 1 was not the reactor involved in the 1979 accident. It was shut in 2019 for operating losses.

Emily Martin

That detail always surprises people. A plant that closed because it lost money is being reopened because a software company will pay for the power. That's a market signal as much as an engineering story.

Zach Martin

It's the whole story in miniature. The customer base for nuclear is changing. For decades the plants had to compete against cheap gas on a wholesale market. Now there's a buyer who values firm, carbon free power at a premium and will sign for twenty years.

Emily Martin

Let me test the premium idea, though. Is this durable, or is it a fashion? Big Tech has made sustainability promises before. If the promises soften, the premium might too.

Zach Martin

Possible, and I won't pretend otherwise. But look at what these contracts actually do. They're long, they're financial commitments, and in some cases like Meta's fuel prepayments and uprate funding, they're cash going out the door before a single kilowatt hour is delivered. That's different from a press release about aspirations. A twenty year PPA assumes AI electricity demand persists for decades, and the buyers are betting their balance sheets on it. They could be wrong. But they're not just talking.

Emily Martin

Okay. So we've got the problem and we've got the buyers. Zach, before we move to fuel, give me the why now in one clean pass. Because nuclear has been available for seventy years. What's different?

Zach Martin

Four things. First, density and reliability. Nuclear is energy dense, it runs around the clock, it uses comparatively little land, and existing reactors can run for decades. Operating plants make power without the direct carbon emissions of fossil generation. Second, a buyer with money and urgency, which we just covered. Third, policy. In May 2025 the administration issued four executive orders on nuclear, including one aimed at compressing licensing, with review timelines in the range of eighteen months for new reactor applications. And the research notes a standard design approval for NuScale's US460 plant, a 462 megawatt design, coming two months ahead of schedule. Regulatory timelines can compress when the politics line up.

Emily Martin

And the fourth?

Zach Martin

Energy security. The United States, China, Europe and India are all competing in AI, semiconductors, robotics, advanced manufacturing, defense, and the electricity to power all of it. Energy has become a strategic technology asset. If you can't generate firm power at scale, you can't run the compute, and if you can't run the compute you don't lead in AI. That framing is why you see bipartisan support for nuclear in a political environment where almost nothing is bipartisan.

Emily Martin

Though I'd add a small caution. Energy security is also a reason nuclear fuel matters so much, because a lot of the fuel supply has historically run through Russia. Which is the perfect hinge into our next section.

Zach Martin

It is. The new reactors, the restarts, the life extensions, and the SMR fleets all need fuel. And that's where the renaissance gets less glamorous and more interesting.