Background
<cite index="19-13">Fractile was founded in 2022 by Walter Goodwin, then a PhD student at the University of Oxford's Robotics Institute, on the thesis that inference speed — not raw training power — would become the industry's defining bottleneck.</cite> Inference is the operational phase of artificial intelligence (AI) in which a trained model processes user queries and generates outputs, as distinct from the computationally intensive training phase.
Architecture
<cite index="11-6">Unlike standard chips that shuttle data between the processor and separate memory modules, Fractile's "memory-compute fusion" architecture keeps data directly on the chip using static random-access memory (SRAM), which does not need to be refreshed.</cite> <cite index="5-2">The company has licensed the Andes AX45MPV RISC-V vector processor — a 64-bit, in-order, dual-issue core with a 1,024-bit Vector Processing Unit and high-bandwidth vector local memory — incorporating it into a data center AI inference accelerator that executes 99.99 percent of the operations needed to run model inference directly in on-chip SRAM.</cite> <cite index="11-7">According to the company, this method can run large language models (LLMs) up to a hundred times faster than existing hardware while lowering operational costs by 90%.</cite> These performance claims have not been independently verified in production.
Funding Round
<cite index="17-2,17-3">Fractile is in advanced talks to notch a valuation more than six times higher than the price it landed in May, raising funds at a pre-money valuation of $6.5 billion and expecting to bring in about $600 million in the round.</cite> <cite index="10-6">That figure includes some money invested at a lower valuation.</cite> <cite index="20-2">Redpoint Ventures and Lightspeed Venture Partners are set to co-lead, with Thrive Capital and Founders Fund also expected to participate.</cite> <cite index="10-8,10-9">The round has not closed and details could change; Fractile and Anthropic both declined to comment.</cite>
<cite index="18-2,18-3">Fractile previously raised a $220 million financing round three months ago, led by Accel, Founders Fund, and Factorial Funds, which gave the British startup a valuation of around $1 billion.</cite> <cite index="1-12">Earlier backers include Kindred Capital, the NATO Innovation Fund, and Oxford Science Enterprises.</cite>
Anthropic Supply Agreement
<cite index="18-5,18-6">The startup has reached an initial deal to sell roughly $250 million of its chips to Anthropic, with the intention to expand that contract in the future; those chips are not expected to be ready for use until 2027.</cite> <cite index="13-3,13-4">Anthropic already holds compute agreements with Google (TPUs), Amazon (Trainium), and Broadcom; adding Fractile would make it Anthropic's fourth major chip supplier.</cite> The agreement is an early commitment rather than a live supply line, and represents institutional validation of Fractile's architecture prior to any production shipment.
Competitive Landscape
<cite index="10-11,10-12,10-13">Fractile is one of several newcomers designing chips tuned for AI inference, trying to claim parts of a market dominated by Nvidia; challengers include Cerebras Systems, the wafer-scale chipmaker, and venture-backed startups such as Etched.</cite> <cite index="10-15,10-16">Etched raised $700 million last week at a $21 billion valuation, while Groq closed a round this month at $3.5 billion, with Nvidia itself joining.</cite> <cite index="13-9">Fractile's team includes engineers from Graphcore, Nvidia, and Imagination Technologies.</cite>
<cite index="6-8">The round ranks in the top 1% of Series C deals for UK chip companies by size, based on a sample of 81.</cite> Fractile's chips are targeted for deployment readiness in 2027, leaving the company pre-revenue at the time of the raise.