Oxford-based AI chip startup Fractile is on track to raise approximately $600 million at a $6.5 billion pre-money valuation, representing a more than six-fold jump from the $1 billion valuation the company achieved just three months ago. The valuation surge follows Fractile’s announcement of an initial $250 million chip procurement agreement with Anthropic, marking one of the largest known deals for a specialized AI inference chip maker.
The preliminary agreement with Anthropic signals growing demand for alternative inference hardware beyond NVIDIA’s dominant position in the market. Fractile’s SRAM-based inference chips aim to deliver faster and more energy-efficient processing for large language model workloads compared to traditional GPU architectures. While the chips aren’t expected to reach production until 2027, the Anthropic deal provides a significant vote of confidence in the startup’s technology and go-to-market strategy.
The funding round has drawn interest from prominent investors, with Accel, Founders Fund, and Factorial all participating. The deal positions Fractile alongside other well-funded inference chip startups including Etched, Groq, and Cerebras, all racing to capture market share from NVIDIA in the rapidly expanding AI infrastructure space.
Fractile emerged from Oxford University as a spinoff focused on solving the computational bottlenecks that arise when deploying frontier AI models at scale. The company’s approach targets the specific memory and processing requirements of inference workloads, where sequential token generation creates different bottlenecks than the parallel training workloads that originally drove GPU demand.
For Anthropic, the deal represents a strategic push to secure diverse inference capacity as the company scales Claude for enterprise customers. With AI inference costs becoming a significant portion of operational expenditure for model providers, securing dedicated hardware partnerships helps hedge against supply constraints and pricing power held by dominant chip vendors.
The valuation jump underscores investor enthusiasm for specialized AI hardware that can deliver meaningful performance improvements over general-purpose GPUs. As more enterprises deploy AI models in production, the inference chip market is expected to grow substantially, creating opportunities for startups that can differentiate on efficiency, latency, or cost.