AI’s Nuclear Power Dealmakers

Zach Martin

This is K M K T I R Hub Radio, and this is Nuclear Capital, the nuclear renaissance powering the AI economy. I'm Zach Martin. Real assets. Real ideas. A brighter tomorrow.

Zach Martin

Before we begin, a clear note. Every company named in this episode is named for educational and market mapping purposes only. Nothing you hear is a recommendation to buy or sell any security. This is editorial analysis, not investment advice, and you should do your own research or speak with a licensed professional before making any decision with your money.

Zach Martin

Here's a sentence worth sitting with. AI cannot manufacture electricity. A model is software. A GPU is a very expensive piece of silicon. Neither one produces a single watt. Every answer a chatbot gives you, every image, every training run, sits at the end of a chain that starts at a power plant, runs through a high voltage substation, crosses a grid interconnect, and only then reaches a rack of chips. The software world loves to talk about scale as if it were free. The physical world sends an invoice.

Zach Martin

And the invoice has a very specific shape. A hyperscale data center doesn't run nine to five. It runs twenty four hours a day, three hundred sixty five days a year, and an outage costs real money. So the question isn't only how much power. It's what kind. Think of a marathon runner versus a sprinter. Wind and solar are valuable, but they produce when the weather allows, with capacity factors that typically land between twenty and thirty five percent. A well run nuclear plant sits at ninety to ninety five percent or better. That's the difference between a power source you schedule around and one you can schedule your business around. Natural gas carries a big share of the load today, roughly forty percent or more of U.S. data center demand by some estimates. But the biggest buyers have decarbonization commitments. They want power that is firm, carbon free, and doesn't need thousands of acres. A multi reactor small modular campus might occupy on the order of fifty acres, versus thousands of acres of solar or wind plus the transmission to move it. Treat those as orders of magnitude, not guarantees.

Zach Martin

Now here's the part I think the market is still absorbing. The customer has changed. For decades, nuclear got built, or kept alive, through regulated utility rate cases. Regulators and ratepayers decided. Today, some of the most heavily capitalized balance sheets on earth are stepping in as direct partners and long term buyers, because what they're really buying is speed to power. Let's walk through who is doing what.

Zach Martin

Start with Google and Kairos Power. Their master agreement aims for up to 500 megawatts of advanced nuclear by 2035, with the first project targeted for 2030. The first deployment is Hermes 2 in Oak Ridge, Tennessee, where Kairos broke ground in April 2026. The plant is designed to deliver up to 50 megawatts to the Tennessee Valley Authority, which serves the grid behind Google data centers in Tennessee and Alabama. According to the August 2025 announcement, TVA is the first U.S. utility to sign a power purchase agreement for electricity from an advanced Gen IV reactor. The design uses a fluoride salt coolant and TRISO fuel, and Kairos raised Hermes 2 output from 28 megawatts to 50 from a single reactor, with operations scheduled to begin in 2030. Google's own energy executive called for smart, firm energy sources. And TVA's chief executive put it simply: energy security is national security.

Zach Martin

Amazon came at it from the reactor and fuel side. In 2024 it anchored a financing round of roughly five hundred million dollars for X-energy, and the two set a target of more than five gigawatts of small modular reactor capacity by 2039. The flagship project is with Energy Northwest in central Washington, using X-energy's Xe 100 high temperature gas cooled reactor. It starts as four units and 320 megawatts, with the option to grow to twelve units and 960 megawatts, targeted for the early 2030s.

Zach Martin

Then there's the near term playbook. Around September 30, 2026, Constellation announced a long term power purchase agreement with Amazon tied to Calvert Cliffs in Lusby, Maryland, reported as a twenty year deal. It covers 690 megawatts, including a 190 megawatt extended power uprate, plus more than three billion dollars of infrastructure investment at the site, and it supports the relicensing effort. The power keeps flowing into the PJM regional grid. This is not a data center bolted to the plant. If you caught my earlier Calvert Cliffs analysis, this is the point: an uprate on a plant that already exists beats a greenfield build to market. That's the commercial playbook for roughly 2026 to 2030. Everything else I'm about to describe is the structural bridge to what comes after.

Zach Martin

And then Meta, which went the portfolio route. In January 2026, following a request for proposals, it announced agreements with Vistra, TerraPower, and Oklo, supporting up to 6.6 gigawatts of new and existing nuclear capacity by 2035. Three pieces. Vistra: twenty year agreements covering more than 2.1 gigawatts from the operating Perry and Davis Besse plants in Ohio, plus power from uprates at those two plants and at Beaver Valley in Pennsylvania, about 433 megawatts of new capacity arriving by the early 2030s. TerraPower: funding for two Natrium sodium fast reactor units, up to 690 megawatts as early as 2032, with rights to up to six more, 2.1 gigawatts targeted by 2035. All together that's 2.8 gigawatts of baseload plus 1.2 gigawatts of built in storage. Oklo: early funding for a campus in Pike County, Ohio, up to 1.2 gigawatts of Aurora powerhouse reactors, a first phase as early as 2030 and full scale by 2034. Existing plants, uprates, and brand new designs, all in one portfolio. That's how a sophisticated buyer hedges technology risk.

Zach Martin

So why nuclear, and why now? Four reasons. Density: a small footprint producing enormous output. Always on: it runs through the night, the winter, the calm. Longevity: plants operate for decades, and licenses can be renewed. And energy security: electricity is increasingly treated as a strategic commodity. Those are the reasons the buyers themselves give, and they hold up. What they don't tell you is how fast any of this can be built. That's where we turn next.

Zach Martin

Let's go upstream, because a reactor is only as real as its fuel. The chain has many links: uranium mining, conversion, enrichment, fuel fabrication, reactor engineering and heavy components, spent fuel and waste management, and cybersecurity wrapped around all of it. Miss one link and the whole thing stalls. Picture a relay race where every runner has to be at the handoff on time. Many of the new designs need a different fuel, called TRISO, short for tristructural isotropic. Here's the detail I love. Each kernel of uranium is about the size of a poppy seed. It's wrapped in layers of carbon and silicon carbide, then packed into graphite pebbles or compacts. Think of each kernel as its own tiny containment vessel. The design aims to hold fission products inside at temperatures above 1,600 degrees Celsius. That's why Kairos and X-energy both lean on TRISO, and why X-energy is building its TRISO X fabrication facility in Oak Ridge.

Zach Martin

Then there's enrichment. Many advanced reactors need high assay low enriched uranium, called HALEU, enriched between five and twenty percent uranium 235. Historically a lot of that supply has come from abroad, which is why domestic enrichment, such as Centrus Energy's work in Piketon, Ohio, matters so much. A reactor design can be brilliant on paper and still wait on a fuel supply that doesn't exist yet. That's a bottleneck, not a footnote.

Zach Martin

Now the reality check on small modular reactors. The promise is elegant. Build standardized reactors in a factory, ship them, assemble them on site, add units as demand grows, and let the learning curve drive costs down. Think assembly line instead of cathedral. But here's the hard part. A factory only beats a construction site if it runs a lot of units. The learning curve needs a high volume, multi unit production line, and in the West that line doesn't exist yet. So the economics are, today, unproven. Until we see repeated schedules and repeated costs, the SMR is a hypothesis with excellent engineers behind it. Even Google's energy executive framed Hermes 2 as a way to learn lessons that help drive down the cost of future reactors. Notice the word future.

Zach Martin

Let me draw the ecosystem map. Not a shopping list. A map. Merchant operators and utilities: Constellation and Vistra. Uranium: Cameco and Uranium Energy Corp. Enrichment, fuel, and components: Centrus Energy and BWX Technologies. Reactor developers and industrial manufacturers: GE Vernova, Oklo, and NuScale, alongside private players Kairos Power, X-energy, and TerraPower. Different parts of the chain carry very different risks. An operating plant under a long contract is not the same animal as a developer still years from commissioning a reactor. I'm naming these companies so you understand who sits where. Again, they are not recommendations to buy or sell.

Zach Martin

The bull case first. AI creates structural demand that doesn't flinch at price, because the chips are useless without power. Existing plants become more valuable. Contracts with investment grade technology companies give multi decade revenue visibility. SMRs and advanced reactors could open up industrial uses and dedicated campuses. And governments increasingly tie AI leadership directly to grid capacity.

Zach Martin

The bear case deserves equal airtime. Nuclear has a long history of multi billion dollar overruns and regulatory delays. First of a kind reactors carry first of a kind costs. Fuel bottlenecks, especially HALEU, are real. AI hardware and algorithms keep getting more efficient, which could shrink power forecasts. And many public nuclear stocks may be pricing in flawless execution years before a single new reactor is commissioned. A great industry and a great investment are not the same thing. That is the line I want every listener to draw.

Zach Martin

Here's the K M K T scorecard, which is editorial judgment, not a price target. AI power demand: nine out of ten. Existing nuclear fleet: nine. Uranium and fuel supply chain: eight. Small modular reactors: eight for long term potential, but just five for near term execution. Advanced Gen IV reactors: nine for long term promise, five for commercial certainty. Overall theme: eight point five out of ten. Notice the pattern. The demand is nearly certain. The delivery is not. High scores on need, middling scores on execution.

Zach Martin

So time horizon matters. Short term, call it 2026 to 2030, this is about operating plants, uprates like Calvert Cliffs and the Vistra plants, license renewals, uranium and enrichment contracting, and the power purchase agreements that make it all bankable. Long term, 2030 to 2035 and beyond, it's factory built SMRs, domestic HALEU output, dedicated nuclear campuses for hyperscale data centers, and Gen IV fast and salt reactors. Coach's rule: don't confuse the sprint with the marathon. The first leg rewards what already exists. The second rewards patience and proof.

Zach Martin

If you want a practical way to follow this without the hype, watch physical milestones, not headlines. Does Hermes 2 get built and connected for 2030? Does TerraPower's first Natrium plant in Wyoming reach completion, which the company expects in 2030? Do uprates at Calvert Cliffs and the Vistra plants come online as promised? Does domestic enrichment deliver fuel in volume? Each one is a checkpoint that tells you whether the long term story is earning its way. Announcements are cheap. Megawatts delivered are not. A press release has never once powered a data center.

Zach Martin

Here's the idea I'd leave with you. The first AI trade was chips and software. The second is infrastructure: the physical, slow, capital hungry world of power plants, fuel, and wires. It doesn't move at the speed of a software release. It moves at the speed of concrete, licenses, and trust. AI runs on electricity. And the future may run on nuclear.

Zach Martin

Thanks for listening. As a reminder, companies named today are not buy or sell recommendations. Zach Martin, Nuclear Capital, an exclusive of K M K T I R Hub Radio. Real assets. Real ideas. A brighter tomorrow.