Energy

A 10 Kilowatt Reactor Just Raised a Billion Dollars

Sequoia led a 1 billion dollar round for Valar Atomics at a 6 billion valuation. The reactor behind it reached 10 kilowatts thermal in June, and the first commercial site planned with Nvidia is 30 megawatts.

EnergyBy Servola Tech Desk2026-08-034 min read

AI-assisted, edited by humans. | Editor-in-Chief: Leon Soliman | Editorial standards

Key takeaways

Valar Atomics reached 10 kilowatts thermal on 22 June against its own 100 kilowatt test design, and the first commercial facility it plans with Nvidia is 30 megawatts, so the demonstrated output is roughly three thousand times smaller than the first real site. Nuclear compute belongs in a 2030 plan, not a 2027 budget.
The valuation tripled from 2 billion to 6 billion dollars in the same period, while no small modular reactor operates commercially anywhere outside China and Russia. The capital is pricing a manufacturing thesis, not delivered electricity.
The figure that is usable now is water, not power: Nvidia puts the design's consumption at close to zero against 2.6 million gallons per megawatt per year, about 9.8 million litres, for conventional cooling. That number can be checked at planning stage, which is more than can be said for the megawatts.
Of the 1.2 billion dollars the company announced, 1 billion is equity and 200 million is a credit facility. Debt raised against a fleet that has not been manufactured is the line to watch, because lenders price manufacturing risk on a schedule that equity does not.
If a supplier quotes you nuclear-backed capacity before 2030, ask which reactor, at what output, and under whose licence. The pilot programme that produced this milestone is a Department of Energy authorisation, not a commercial licence from the Nuclear Regulatory Commission.

What the reactor has actually produced

On 18 June the Department of Energy announced that Valar Atomics' Ward 250 had reached zero-power fueled criticality at the Utah San Rafael Energy Lab outside Orangeville, in Emery County. It was the second reactor to go critical under the department's Reactor Pilot Program, after Antares Nuclear's Mark-0 at Idaho National Laboratory, and the first to do it outside a national laboratory. Energy Secretary Chris Wright called it a revolutionary moment for advanced nuclear, and listed the airlift of a small reactor aboard a US military C-17 alongside the criticality test.

Criticality is not power. It means the reactor can hold a controlled chain reaction, which it must do before it generates anything, and Valar then began the ascent. By 22 June the company reported 10 kilowatts thermal. The unit is a TRISO-fuelled, helium-cooled high-temperature gas reactor rated at 100 kilowatts thermal, so 10 kilowatts is a tenth of what this particular machine was built to make. Chief executive Isaiah Taylor was direct about the intent: the reactor was built to make power, and that is where they are headed. Nothing in that sequence is slow by the standards of the industry. It is simply small, and the size is the part that matters for planning.

Three thousand times separates the demonstration from the first site

The commercial application Valar and Nvidia have described is a 30 megawatt nuclear-powered AI facility in Utah. Thirty megawatts is thirty thousand kilowatts. Set against the 10 kilowatts thermal demonstrated in June, the first commercial site is on the order of three thousand times larger, and the true gap is wider still, because the demonstrated figure is thermal while a facility rating is electrical and a gas reactor loses a substantial share of its heat in conversion.

This is an argument about scheduling, not about credibility. Valar has done in months what most nuclear programmes measure in years, and the DOE put its name to it. But the round was raised at 6 billion dollars, three times the 2 billion valuation of the previous round, in a market where no small modular reactor is in commercial service anywhere outside China and Russia. Taylor framed the use of funds precisely, and the framing is the tell: one reactor can be built as a project, a fleet has to be manufactured. The money is buying a factory thesis. A factory thesis is a five to ten year proposition even when it works.

The deliverable that lands first is a water figure

Read the announcement as a data-centre operator and one number is immediately usable, and it is not a number about electricity. Nvidia's stated case for the closed-loop helium design is that it takes facility water consumption from 2.6 million gallons per megawatt per year, roughly 9.8 million litres, to nearly zero. Cooling water, not generation, is what has been blocking permits in the places where compute wants to go.

That distinction has practical value today. A siting team can put a water figure into a planning application, an environmental assessment or a conversation with a regional authority long before a single reactor is licensed to sell power to anybody. The megawatts are a 2030s conversation. The water is a 2026 conversation, and it is the part of this announcement a European operator can act on, because the constraint it addresses is the same one that already governs where a data centre may be built and how much of a local supply it is permitted to draw.

What goes in the 2027 budget, and what does not

Treat the 1.2 billion dollar figure the company published with the same care you would give any headline raise, because it is two instruments. One billion is Sequoia-led equity, with Atreides Management, Point72 and Snowpoint Ventures alongside and Sequoia's Shaun Maguire taking a board seat. The remaining 200 million is a credit facility. Debt against a fleet that does not exist yet is the more informative half: lenders attach schedules and covenants to manufacturing risk in a way equity investors do not, so the terms of that facility will say more about the real production timetable than the valuation does.

For anyone building a capacity plan, the instruction is narrow. Do not put nuclear-backed power into a 2027 or 2028 budget line on the strength of this round, and do ask any supplier offering it which reactor, at what output, and under whose licence, since the Reactor Pilot Program is a Department of Energy authorisation to demonstrate and commercial sale runs through the Nuclear Regulatory Commission on a separate track. Do, however, put the water assumption into your siting work now. It is the only part of this that is available to you in the planning cycle you are actually in.

Frequently asked questions

Is 10 kilowatts a failure or a normal milestone?

A normal and fast milestone. Reactors go critical at zero power first, then ascend, and Valar reached criticality and 10 kilowatts thermal within days. The point is not that the number is disappointing; it is that a 30 megawatt facility is three thousand times larger, so the milestone does not shorten a commercial timetable.

Does the DOE pilot mean the reactor is licensed to sell power?

No. The Reactor Pilot Program is a Department of Energy authorisation to build and demonstrate first-of-a-kind designs. Selling power commercially runs through the Nuclear Regulatory Commission, which is a separate process on its own schedule. Treat a pilot milestone as evidence about engineering, not about permission.

What should a European operator take from a US funding round?

The water arithmetic, chiefly. The claim that closed-loop cooling removes roughly 9.8 million litres per megawatt per year is testable against your own siting constraints and applies wherever water availability, not generation, is what limits where you may build. The generation timetable is American, tied to American regulators, and will not arrive in a European capacity plan this decade.

Advanced nuclear is being funded years ahead of what it can deliver, and that is a rational way to finance manufacturing; factories have to be paid for before they produce. The mistake available to an operator is to read a valuation as a delivery date. Ten kilowatts is a real achievement and a poor basis for a power budget. Take the water number, which is checkable now, leave the megawatts in the decade they belong to, and revisit when a reactor has sold electricity to a customer under a commercial licence.

Valar AtomicsSequoia CapitalNvidiasmall modular reactorDepartment of EnergyReactor Pilot Programdata centre waterNuclear Regulatory CommissionAI powerUtah

Britain Bought a Chip You Cannot Buy Until 2027

The UK Sovereign AI fund took equity in Olix as it raised 312 million dollars at a 3.3 billion valuation. The DX-1 reaches its first customers in the second half of 2027, two budget cycles away.

4 min read

Infrastructure

Compute Doubles Faster Than Its Price Falls

Epoch AI publishes at least three doubling times for AI compute. The one in every headline measures something else, and the curve that reaches your invoice is moving about three times slower.

6 min read

Law & Regulation

The EA Buyout Now Hinges on Brussels

The largest take-private in history has cleared EA shareholders and US antitrust. The last real gate is the EU - and not classic merger review, but the Foreign Subsidies Regulation, facing its first blockbuster test.

3 min read

← All articles