Fractile

AI inference — the moment a model responds to a query or generates an output — has become the new frontier of costs in the AI ecosystem. As models grow larger and outputs longer, the bottleneck has shifted from training to real-time response. Fractile was founded to address exactly this problem: it develops chips and hardware systems specifically designed for inference, with an architecture optimised to reduce latency and cost-per-token compared to general-purpose chips. Based in London, the startup raised $220 million in a Series B led by Accel, Factorial Funds, and Founders Fund, with participation from Conviction, Gigascale, O1A, Felicis, Buckley Ventures, and 8VC. This is one of the largest rounds ever closed by a European semiconductor startup, signalling growing global investor attention towards AI hardware as critical infrastructure. Fractile's thesis is that agentic AI systems — those that reason across multiple steps, generate long responses, or handle parallel streams — require hardware architectures built from scratch for this use case. Current chips, originally designed for training or general-purpose computing, are not optimised for agentic inference and generate operating costs that slow enterprise adoption. The timing is strategic: in 2026, agentic AI is no longer a future concept but a production reality deployed by thousands of companies. Demand for efficient inference hardware is set to grow rapidly, and Fractile positions itself as a European alternative to dominant American players in the sector.