London-based chip designer Fractile has raised $720 million in a Series C round, betting that memory bandwidth, rather than raw computing power, is the main constraint holding back AI performance.
The funding values the four-year-old company at $7 billion, according to the , which cited people familiar with the deal.
Thrive Capital, Founders Fund, Lightspeed, and Redpoint led the round. Other investors include Alphabet’s CapitalG, DST Global, BlackRock, Bain Capital Ventures, T. Rowe Price, Shopify Ventures, Balderton, the UK’s Sovereign AI fund, and the British Business Bank.
The valuation is a sharp jump from May, when Fractile raised around $220 million at a valuation of roughly $1 billion. The new figure puts it among the UK’s most valuable startups, alongside companies such as Revolut.
The also reports that Anthropic has expressed interest in becoming an early customer for Fractile’s chips, citing people familiar with the discussions.
Founder Walter Goodwin’s argument starts with a growing imbalance between computing power and memory bandwidth.
Over the past 25 years, chip performance has increased roughly a millionfold, while memory bandwidth has grown only about 40 times. That leaves a 25,000-fold gap between the amount of data processors can calculate and how quickly they can access it.
Goodwin told the FT that today’s frontier models are “drinking through a straw”.
Fractile’s proposed solution combines custom DRAM and logic in a three-dimensional stack. The company says this architecture can deliver substantially higher bandwidth than high-bandwidth memory (HBM), while retaining enough capacity to run large models with long context windows. Goodwin claims it could make inference, the process of generating responses to user queries, up to 10 times faster.
Those performance gains have yet to be demonstrated in a production chip. Fractile has not released a commercial product, and its first chip is planned for next year. For now, the $7 billion valuation is based on its technology and prototypes rather than a product in the market.
The company is also trying to avoid a growing constraint in the AI supply chain. Many competing chipmakers rely on HBM to move data quickly between memory and processors.
The technology is expensive, and demand from AI companies is putting pressure on suppliers such as SK Hynix and Micron.
Fractile is entering an increasingly crowded market. Nvidia acquired Groq last year and launched its own inference chip in March, while startups including Etched, MatX, and fellow British company Olix are developing alternatives.
The opportunity, according to Thrive Capital’s Philip Clark, is an inference market that could soon become “at least as large as the market Nvidia is addressing today”.
He sees particular potential in AI agents, which need to run continuously for hours or days rather than respond to a single prompt.
The funding is also a significant moment for the UK’s effort to build domestic AI infrastructure. Goodwin described the government’s Sovereign AI fund as a driving force behind developing frontier AI technology from London.
Fractile also has offices in the US and Taiwan, however, and is building for a global market. Its next challenge is to turn its prototypes into working chips that can deliver the performance it has promised.