Fractile: The future of AI inference

AI is evolving from short answers to sustained work.

The next generation of applications will involve models that reason, search, code, test, and iterate across long sequences of work. That changes the demands on the infrastructure beneath them. As models spend more time thinking at runtime, inference becomes one of the most important constraints in AI.

For much of the AI boom, training has dominated the infrastructure conversation. But inference is where AI capability turns into useful output, and where speed, cost, and latency determine what can actually be built.

Fractile, founded in 2022 by Walter Goodwin, is building chips and systems designed specifically for the inference workloads now emerging. The team is rethinking compute, memory bandwidth, and software together to make AI systems dramatically faster and more efficient.

Why inference matters now

For many AI products, the user experience is already shaped by inference. A faster model feels more capable. Lower-cost inference means models can be used more often. When systems can generate more output without increasing cost or latency, they can attempt harder work.

This becomes even more important as AI moves into agentic coding, scientific discovery, enterprise automation, and long-horizon research. These use cases require models to keep working through the problem - exploring possibilities, checking outputs, and refining the answer.

Fractile’s view is that today’s systems were not designed for this world. Solving the inference bottleneck requires more than incremental improvement. It requires rethinking the stack from the start.

Built from first principles

What stood out to us about Fractile is the ambition to build across the stack. This is a hard company to build: it requires deep expertise across AI research, chip architecture, systems design, software, and manufacturing.

Walter and the Fractile team bring exactly that kind of depth. Walter, who completed a PhD in AI at Oxford, identified the need for specialist inference hardware early and has stayed focused as the market has evolved. Around him, Fractile has assembled an exceptional group of engineers and researchers with the technical range needed across both hardware and software to take on one of the hardest problems in AI infrastructure.

That combination of technical ambition, commercial clarity, and long-term conviction is rare. It is also what makes Fractile so exciting.

That’s why we’re delighted to co-lead Fractile’s Series B. The UK has a long history of world-class talent in AI and semiconductors. Fractile brings those strengths together at exactly the time the market needs new answers.

If AI is going to move from impressive demos to systems that can tackle the hardest problems, inference has to become faster, cheaper, and more scalable. Walter and the Fractile team are building for that future, and we’re excited to partner with them.