Fractile joins UK push to build faster, cheaper AI chips

In short: More UK companies are building new computer chips to run AI faster and with less electricity, and Fractile is one of the latest to claim big gains.

What's going on

Fractile, a UK-based AI hardware startup, is developing a new chip architecture aimed at AI inference. AI inference is the part where an AI system gives you an answer after it has already been trained (like a calculator giving you a result, not learning math).

Fractile says its approach can deliver up to 25 times faster performance and cut costs by up to 10 times compared to systems built around GPUs. A GPU is a common type of chip used for AI today, and it is powerful but often expensive to buy and run.

Fractile is part of a wider UK trend. Several newer firms are trying to reduce the energy and cost problems that come with running AI on GPUs at scale. Literal Labs, founded in Cambridge in 2024, says its logic-based tools for training AI models are far faster and more energy-efficient than neural networks on GPUs. Vertical Semiconductor, also founded in Cambridge in 2024, is working on Vertical GaN semiconductors, which are materials meant to handle power more efficiently. Spectral Compute, founded in London in 2018, builds accelerated computing tools meant to work across different hardware.

The Financial Times also points to the size of the UK AI sector, valued at £72.3 billion, as a backdrop for this growing focus on custom AI hardware.

What to watch

The key question is whether these performance and cost claims hold up in real-world use, and whether big AI buyers adopt these chips. It will also matter how quickly these companies can manufacture at scale, since making chips in volume is often harder than designing them.

Source: Financial Times