London-based semiconductor startup Fractile announced on May 13, 2026, that it had raised $220 million in a Series B funding round to accelerate the development and commercialization of its forthcoming AI (Artificial Intelligence) inference chips. <cite index="1-2,1-3">The round was co-led by Accel, Factorial Funds, and Peter Thiel's Founders Fund</cite>, with <cite index="1-6">participation from Conviction, Gigascale, O1A, Felicis, Buckley Ventures, and 8VC</cite>. <cite index="9-9">Existing backers Kindred Capital, the NATO Innovation Fund, and Oxford Science Enterprises, which co-led Fractile's $15 million seed in July 2024, are part of the round</cite>.
Technical approach
<cite index="1-8,1-9">Fractile was founded in 2022 by Dr. Walter Goodwin, a then PhD student at the University of Oxford's Robotics Institute. The company is developing chips that use in-memory compute, an approach that allows processors to run calculations directly in computer memory.</cite> The architecture is intended to address memory-bandwidth limits that constrain throughput on conventional GPUs (Graphics Processing Units). <cite index="9-11">Conventional AI accelerators, including Nvidia's H- and B-series GPUs, separate the compute die from high-bandwidth memory</cite>, a bottleneck Fractile says becomes acute as model outputs grow.
Fractile frames the problem in terms of token economics. <cite index="10-11">The company said current systems typically generate around 40 tokens per second for such large workloads, while more advanced use cases may require output speeds closer to 1,200 tokens per second</cite>. According to founder Walter Goodwin, faster inference could compress workloads that currently take weeks or months into hours or days.
Customer interest and market position
<cite index="1-11">Earlier this month, a report from The Information claimed that generative AI company Anthropic had held discussions with Fractile regarding the purchase of the startup's inference chips when the hardware becomes available in 2027.</cite> <cite index="9-14,9-15">If the relationship formalises, Fractile would become Anthropic's fourth named compute supplier alongside Nvidia, Google's TPUs, and Amazon's Trainium and Inferentia parts. Anthropic has separately been exploring building its own custom AI chips, but the Fractile track suggests it is still pursuing a multi-supplier hedge.</cite>
The round arrives within a broader thesis among investors that inference is structurally distinct from training. <cite index="9-18">The argument is that training will continue to require the largest, most exotic systems and that Nvidia's CUDA moat is strongest there, while inference, the workload that actually consumes most of the dollars once a model is deployed, rewards specialised</cite> silicon. That logic has also drawn capital to U.S. competitors including Groq, Cerebras, and SambaNova, and prompted hyperscalers to commission custom parts.
UK context and operations
The financing follows a previously announced UK expansion. <cite index="1-10">In February 2026, the startup announced plans to invest £100 million ($135m) to bolster its UK operations over the next three years, with the expansion set to include the growth of its existing sites in London and Bristol, and the creation of a new hardware engineering facility in the latter city.</cite> <cite index="9-22">The team has drawn engineers from Graphcore, Nvidia, and Imagination Technologies, and is building its software stack alongside the silicon.</cite> <cite index="5-12">The company is hiring across London, Bristol, San Francisco, and Taipei.</cite>
Fractile has not disclosed a foundry partner, published independent benchmarks, or set a firm commercial-availability date beyond the 2027 window referenced in press reports. The company's claims of in-memory compute advantages on a watts-per-token basis remain unverified by third parties, a caveat that applies broadly to early-stage entrants in the inference-accelerator market.