A Nuclear Startup Just Raised $1 Billion to Power AI — but the Clock Is Running

A Nuclear Startup Just Raised $1 Billion to Power AI — but the Clock Is Running

Let me tell you something that happened this morning that most of the industry is going to misread completely. Valar Atomics — a nuclear startup out of Utah that didn't even exist three years ago — confirmed it raised $1 billion in equity, led by Sequoia, at a valuation Bloomberg pegs around $6 billion. Shaun Maguire, a Sequoia partner, is joining the board. On top of the equity, they locked in a $200 million line of credit from Erebor and other banks.

I've been running hosting infrastructure for over a decade, and I've watched every power story in this industry get sold as the thing that finally fixes the AI energy problem. Gas turbines. Grid upgrades. Closed-loop cooling. Solar farms. And now, the nuclear era. But here's the thing nobody in the comment section is going to tell you: this is not a story about whether nuclear can power AI. It's a story about time. And time is the one thing the AI buildout does not have.

The $1 Billion Bet — What Sequoia Actually Bought

Let me be clear about what happened, because the headlines are going to oversimplify it. Valar Atomics is building small modular nuclear reactors — SMRs. Factory-built, miniaturized power plants that are supposed to be cheaper and faster to stand up than the multi-billion-dollar behemoths we've been building since the 1970s. The company's founder and CEO, Isaiah Taylor, started it in 2023 when he was 24, with the pitch that nuclear reactors ought to be built on a factory line, not in the middle of a field over a decade.

The investors writing the checks are not your average tech angels. Sequoia is the firm that backed Apple, Google, Nvidia, and OpenAI. Point72 is in there. Conviction is in there. Riot Ventures, Valor Equity Partners, Atreides Management, HOF Capital — a syndicate that reads like the AI infrastructure power index. And they're not alone. Antares raised $470 million for its microreactor. X-energy raised $1 billion through its IPO. The money is pouring into this sector like it's 2021 all over again — except this time the product is not a social app, it's a fission reactor.

Now, the company line is that this funding gets them to the next milestone: manufacturing fleets of SMRs. They say it took two years to complete the NOVA core, seven months to take Ward 250 critical, and with each reactor built, the tick rate gets smaller — until they're producing tens, then hundreds, then thousands of reactors per year. That's the pitch. Thousands of reactors a year, coming off a factory line, like they're toasters with a half-life.

Ward 250 — The Reactor That Actually Went Critical

Here's the part that separates Valar from the vaporware crowd: they actually built the thing. On June 18, Ward 250 — their Gen IV TRISO-fueled modular high-temperature gas reactor — achieved self-sustaining criticality at the San Rafael Energy Lab in Emery County, Utah. That's the point where the fission chain reaction keeps itself going. It was the second advanced reactor to go critical under the Department of Energy's Reactor Pilot Program, and the first DOE-authorized reactor built and operated outside the national laboratory system.

I want you to sit with that for a second, because it's genuinely unprecedented. A startup — not Los Alamos, not INL, not Westinghouse — built a reactor outside the national lab system and got it to criticality, under Executive Order 14301, which directed DOE to hit criticality on at least three advanced reactors by July 4, 2026. They beat the deadline. The reactor is now generating 100 kilowatts of electricity, which makes it the first time in history a startup has produced nuclear power. When Taylor says "Valar Atomics is the company that is done waiting for someone else to unleash the power of the atom," he's not being cute. He did it.

And the fuel story is the part that makes engineers' eyes go wide. The fuel in Ward 250, if concentrated, would form a five-inch cube. Taylor says that cube holds enough energy to send a fully loaded Saturn V to the moon twelve times, or drive your car around the entire Earth 7,000 times. TRISO fuel is the same stuff that's been described as the most robust nuclear fuel ever developed — each particle is a ceramic-coated uranium kernel that can survive temperatures that would melt most reactor designs. The company says the fuel load could keep the reactor running for 165 years. That's not a power plant. That's a battery for a civilization.

The Nvidia Deal — Waterless AI Factories and the 30 Megawatt Promise

But a reactor sitting in the Utah desert is not the same thing as AI infrastructure. That's where the Nvidia deal comes in, and this is the part that should genuinely interest every one of you running servers. In June, Valar demonstrated Ward 250 powering an Nvidia Blackwell system — and announced a collaboration to develop a waterless 30-megawatt AI factory in Utah. If it proceeds, it's the first time advanced nuclear directly powers AI.

I wrote about the AI water crisis a couple of weeks ago — 264 billion gallons a year, drought-stricken regions fighting over every drop, communities demanding disclosure. So pay attention to this design. The Ward 250 uses a helium cooling system, and Nvidia's new DSX factory design uses direct liquid cooling where water circulates at 113 degrees Fahrenheit — eliminating the power-hungry chillers that eat so much of a data center's load. The proposed facility is 300 times smaller than the approved Stratos data center in Box Elder County, which tells you exactly how the industry is thinking now: not bigger, but self-contained. The reactor, the cooling, the compute, all on one site, with almost no water draw.

Valar's head of projects, Max Ukropina, framed it as a promise to communities: high-skill jobs in reactor operations and data center engineering, tax revenue, supply chains — without new demands on local water supplies. And look, I get the appeal. After a year of watching communities fight data centers tooth and nail over power and water, a facility that promises both is exactly what the industry wants to hear. Nvidia's Levi Patterson shook hands on that stage in Orangeville and it was the handshake of an industry desperate for a path forward.

The Counter-Argument — and the Part That Keeps Me Up at Night

Now let me tell you the part that the press release doesn't. The reactor works. The physics is proven. But commercial power is a completely different game, and there are three walls between Ward 250 and your colo provider's power meter.

First, the Nuclear Regulatory Commission has to license this thing. The NRC chairman, Ho Nieh, said the commission is hoping to shorten licensing review times to under 18 months. Hoping. That's not a commitment, and even 18 months is an eternity in an industry that needs power yesterday. Second, the fuel supply chain — HALEU, high-assay low-enriched uranium — is still being built in this country. Centrus just signed a $900 million DOE contract to pivot its HALEU cascade to commercial operation, but that's a supply chain in its infancy. And third, the manufacturing story — "thousands of reactors per year" — is a claim about a factory line that doesn't exist yet. The tick rate gets smaller with each reactor. But they've built one. One.

Here's the uncomfortable truth: the AI buildout's power problem is a now problem, and nuclear is a later solution. Every hyperscaler has figured this out — Microsoft is betting on Three Mile Island and its $60 million Genesis mission commitment, Google is backing Kairos, Amazon is in on X-energy. They're all buying nuclear because they know the grid can't deliver. But every single one of those reactors is a 2030s story, and the data centers going up right now are being powered by gas turbines and grid upgrades that communities are already fighting in court. The nuclear industry just raised a billion dollars to build the future. The present is still burning diesel.

What This Actually Means for Independent Hosting Providers

So what do you do with this if you're running an independent hosting business? Because I know that's who's reading this, and I know the AI capex headlines make you feel like a spectator.

First, don't plan your capacity around nuclear power. I'm not saying it won't happen — I'm saying it's a decade out, and your customers need power next quarter. If you're in a region betting on SMRs to rescue the grid, treat it as a bonus, not a plan. The PJM-style auction shortfalls and grid interconnection queues are your real constraint, and they're not going anywhere.

Second, watch the waterless cooling designs. The DSX-style direct liquid cooling at 113 degrees — no chillers, minimal water — is the most transferable idea in this whole story. When you spec your next facility or colo contract, ask about water usage effectiveness and chiller loads. The providers who crack waterless operation are going to have a pricing advantage in drought-prone markets, and that advantage is coming sooner than any reactor.

Third, treat nuclear equity raises as a signal, not a solution. When Sequoia puts a billion dollars into reactors, it tells you the hyperscalers have hit the wall on conventional power — that's the real news. It means grid-constrained regions stay constrained, colo pricing stays elevated, and the premium on sites with firm power contracts keeps climbing. That's your market signal. Position your business accordingly: lock power contracts early, get your permitting ducks in a row now, and be ready to serve customers who are fleeing regions where the buildout is stuck.

The Structural Reality — This Isn't a Technology Story, It's a Calendar Story

Here's the way I see it after a decade of watching this industry. The nuclear-for-AI story is real, but it's running on a fundamentally different clock than the AI buildout. The hyperscalers are spending hundreds of billions on compute today and borrowing against tomorrow to do it. The nuclear startups are raising billions to deliver power in the 2030s. Both of those things are true at the same time, and neither one fixes the other's timeline.

The uncomfortable parallel is the GPU shortage all over again. Everyone remembers the chip shortage — I wrote about how AI was eating every chip in the supply chain. This is the same pattern, just slower and scarier. Instead of fab lead times, it's NRC licensing. Instead of memory allocation, it's HALEU fuel. The bottleneck moved from the silicon to the atom, and the lead time got longer.

Valar Atomics deserves real credit. They did something no startup has ever done, and they did it fast. But a critical reactor is not a commercial fleet, and a $6 billion valuation is not a power grid. If you're building an AI empire on the promise that the atoms will save you, I'd remind you of one thing: Ward 250 generates 100 kilowatts. A single modest data center rack can pull more than that. The gap between the demonstration and the deployment is the whole story, and that gap is measured in years.

The Bottom Line

Here's what I'd tell you if we were standing at the bar after work. The nuclear money is real, the physics is real, and the reactor is real. I'm not here to tell you nuclear won't power AI — I believe it eventually will. What I'm telling you is that "eventually" is the most expensive word in this industry, and somebody's going to pay for the years in between.

Independent hosting providers have one advantage the hyperscalers don't: we can't afford to wait, so we can't afford to be wrong. While the giants buy 2030s power at 2026 prices, the smart independents are locking in what's available now, building water-smart facilities, and positioning themselves as the providers who can actually deliver. The reactor people are building the future. We're the ones who have to keep the lights on until it gets here.

Watch the licensing calendar. Watch the HALEU supply chain. Watch what the hyperscalers do when the grid keeps saying no. Because the next chapter of this story isn't going to be written in a reactor core — it's going to be written in a permitting office, and the winners are going to be the ones who understood the calendar before the crowd did.

— Allan Ali, Founder

This article was produced with AI-assisted research and editorial support. Reporting is based on sources cited in the article.