Walter Goodwin

Walter Goodwin founded Fractile in 2022, after completing a PhD in robotics, to build processors for running trained AI models. Oxford Science Enterprises has backed the company since its founding. It works from central London and Bristol and is hiring in San Francisco and Taipei.

A bet on inference speed

Fractile’s argument is that the time a model takes to produce an answer has become the main limit on what frontier AI can do. Models working on hard problems can now generate up to 100 million tokens, and at the roughly 40 tokens per second typical on today’s chips, one such output takes about a month. The bottleneck, the company says, is memory bandwidth that has not kept pace on current architectures.

Its answer is a processor in which memory and compute are physically interleaved, designed to serve thousands of tokens per second to thousands of users at once. Fractile says its systems will run the most advanced models up to 25 times faster at a tenth of the cost. “Fractile’s AI accelerator hardware is uniquely equipped to meet the unending growth in demand for increased performance and capacity of AI compute,” Goodwin said in a story published by Oxford Science Enterprises.

From £100m UK plan to a $220m round

In February 2026 the company committed £100m over three years to expand its London and Bristol sites and create a UK industrial hardware engineering facility, as reported by Tech.eu and Computer Weekly. At the time it employed 80 people. The government used the announcement to showcase its AI opportunities plan.

In May Fractile raised $220m (£164m) to get its first chips and systems to customers, in a round led by Accel, Factorial Funds and Founders Fund, with Conviction, Felicis and 8VC among the other investors. In August Bloomberg reported that Fractile had reached an initial agreement to supply about $250m of chips to Anthropic from 2027, and that it was in talks to raise about $600m at a $6.5bn pre-money valuation, according to The Next Web. Both companies declined to comment, and neither has announced a deal.

Why it matters to UK buyers

For businesses, the cost of running AI models day to day is set largely by the hardware underneath. A UK-designed inference chip, built by a team spread across London and Bristol, gives data center operators and cloud providers a potential alternative to the handful of US suppliers that dominate today. The £100m engineering commitment also keeps skilled hardware jobs in the UK.

Goodwin has built one of very few UK companies designing its own AI silicon and raised $220m in 2026 to bring it to market. The first chips in customers’ hands will show whether the design delivers the speed its investors are backing.