Fractile Ltd., a London startup designing chips to run artificial-intelligence models, has raised about 577 million pounds, or roughly $774 million, in a Series C round it has not publicly announced, according to share allotment filings with the U.K.’s Companies House.
The filings, lodged on Oct. 15 and Oct. 28, show three new classes of preferred shares. The largest, Series C-3, accounts for about 392 million pounds at 1,452.35 pounds a share, issued in tranches from Sept. 9 through Oct. 23. Two smaller classes, C-1 and C-2, were sold on Sept. 11 at about 802 pounds and 707 pounds a share, raising roughly 148 million pounds and 37 million pounds. The filings don’t name investors, and the company hasn’t explained why some backers got lower prices.
A fivefold step-up
The top Series C price is about 5.2 times what investors paid for Series B shares in May, when Fractile raised a round reported at $220 million. At that price, the roughly 2.64 million shares in issue are worth about 3.84 billion pounds, a figure that excludes options and any further allotments. Fractile was earlier reported to be seeking about $600 million at a $6.5 billion valuation before new money.
The step-up followed a commercial breakthrough. Anthropic agreed to an initial order of about $250 million of Fractile chips, with both sides interested in expanding it, according to people familiar with the agreement. Fractile has not yet shipped a production chip and expects its hardware to reach data centers in 2027.
Betting on inference
Founded in 2022 by Oxford-trained roboticist Walter Goodwin, Fractile designs chips for inference, the stage at which a trained model produces answers. Its design keeps data in memory placed next to the circuits doing the math, an approach meant to cut the time and energy spent moving data. The company runs engineering teams in London and Bristol and has pledged 100 million pounds over three years to expand in the U.K.
The raise puts Fractile among Europe’s best-funded hardware startups as investors chase alternatives to Nvidia for inference work, following rounds such as General Compute’s $100 million and FuriosaAI’s $620 million this past week.
