Fractile, a chip startup building inference hardware for frontier AI systems, has raised £160 million in funding led by Accel, Factorial Funds and Founders Fund, with participation from Conviction, Gigascale, O1A, Felicis, Buckley Ventures and 8VC alongside existing investors. The funding will be used to accelerate the path to getting Fractile’s first chips and systems into customers’ hands.
Founded in 2022, Fractile develops chips and systems designed to improve the speed of AI inference workloads. Fractile said current frontier AI systems are increasingly constrained by the time required to generate outputs and by the limits of memory bandwidth on existing hardware architectures.
According to Fractile, some AI workloads already require outputs of up to 100 million tokens. At around 40 tokens per second on existing chips, Fractile said generating outputs of that length can take around a month. Fractile said compressing that workload into a day would require output speeds of around 1,200 tokens per second while operating large models over very long contexts.
The business said its systems are designed to tackle the complexity, latency and capacity challenges involved in running frontier AI models over long sequences. Fractile said the workloads enabled by faster inference could support applications including software engineering, drug discovery and materials discovery.
Since founding, Fractile said it has worked across AI research, foundry process innovation and chip micro-architecture development. Fractile is hiring across London, Bristol, San Francisco and Taipei.
