A startup building miniature nuclear reactors raised one billion dollars in a single round. Not for a reactor that works and sells electricity - for a factory that has yet to manufacture them. This is Valar Atomics, and the round is led by Sequoia, with its partner Shaun Maguire joining the company's board.

Alongside the equity, the company secured a credit line of 200 million dollars from Erebor and several other banks. The valuation was not announced officially, but according to Bloomberg it is 6 billion dollars. For comparison: this is a firm whose product is still not in serial production.

What they are building are called small modular reactors - the idea is that the power plant gets assembled in a factory like any other product, instead of being built on site over years. If it works, you get something cheaper and faster than a conventional nuclear plant. If it does not work, you get a very expensive warehouse full of parts.

So far they have somewhat more than a presentation. In June, they say, their Ward 250 reactor powered an Nvidia system with Blackwell chips, and with the same company they agreed to build a 30-megawatt AI facility that runs without water. That is a concrete result, not a slide.

Founder and chief executive Isaiah Taylor explained the pace with numbers: „It took two years to finish the NOVA core. It took seven months for Ward 250 to go critical. With every reactor built, the cadence will shorten until Valar is producing dozens, then hundreds, then thousands of reactors a year.”

Thousands of reactors a year. That is a sentence worth reading twice. And it is exactly what the investors paid for - not for the seven months already behind them, but for the curve that promises it will be faster next time.

Valar is not alone in this race. Antares recently raised 470 million dollars, and X-energy pulled in a billion by going public. Valar's round also drew in Point72, Valor Equity Partners, Atreides Management, Riot Ventures and several more funds.

Money in nuclear power is suddenly everywhere, and that is no accident - AI data centres consume electricity at a pace existing grids cannot withstand, so somebody has to produce it. The question rarely raised in the announcements is who carries the risk if all these billions ultimately yield a few prototypes and one very long queue for permits.