As the AI race increasingly pivots from training ever-larger models to the challenge of running them efficiently, inference has become a critical bottleneck. London-based startup Fractile is targeting this precise layer, developing chips purpose-built for the next wave of AI agents. The company has now closed a $220 million (approximately €187 million) funding round to accelerate that vision.
The investment will fuel the development of novel processor architectures designed to handle the unique demands of agentic AI workloads. Unlike conventional accelerators optimized for training, Fractile’s silicon is engineered for real-time execution — where latency, throughput, and energy efficiency determine whether autonomous agents can operate at scale. While specific investors and timeline were not detailed, the size of the round underscores the market’s conviction that dedicated inference hardware will be essential as models become more capable and companies deploy them in production.
By focusing on this under-addressed segment of the AI stack, Fractile aims to provide the computational foundation needed for the next generation of intelligent systems, positioning itself at the heart of a shift from building models to making them reliably useful in the real world.