UK moves to secure homegrown AI chips

Britain is preparing to use public procurement to support domestic artificial intelligence chipmakers, offering to buy semiconductor equipment from UK-based technology companies as ministers seek to stop promising hardware firms from shifting capital, talent and production plans overseas.

Technology Secretary Liz Kendall is expected to outline the plan at London Tech Week, where the government will position strategic purchases as part of a broader effort to give early-stage chip designers a credible first customer. The proposal is aimed at companies developing processors and systems for AI inference, the fast-expanding market for running trained AI models at lower cost and higher speed.

The move reflects growing concern in Whitehall that the UK has produced world-class technology assets but has too often failed to retain the commercial upside. Arm, founded in Cambridge, remains central to global chip architecture but chose New York for its stock market listing. Graphcore, once regarded as one of the country’s most promising AI hardware companies, was sold to SoftBank after struggling to compete with better-funded rivals. Ministers now want government demand to help anchor the next generation of AI hardware firms in Britain before overseas buyers or investors set the terms.

Kendall’s department has already placed AI compute, semiconductors and sovereign technology capacity at the centre of its growth agenda. The government’s compute roadmap includes investment of up to £2 billion to expand public compute infrastructure, with more than £1 billion earmarked to increase the AI Research Resource twenty-fold by 2030 and up to £750 million for a new national supercomputing service in Edinburgh. The chip purchase plan would add a market-making element to that strategy by linking public-sector demand with domestic hardware development.

The proposed approach also fits with a wider shift towards more interventionist industrial policy. A £500 million sovereign AI fund has begun taking stakes in AI companies, while the Advanced Research and Invention Agency has launched a £50 million AI chip testbed to help start-ups validate designs and attract customers. The government’s argument is that grants alone are not enough; young hardware firms need buyers, testing facilities, patient capital and access to compute infrastructure if they are to scale from laboratory promise to commercial deployment.

Several UK start-ups have drawn investor interest as demand for alternatives to Nvidia’s dominant graphics processing units rises. Fractile, founded in 2022 and based in London, is developing chips and systems designed to accelerate AI inference by bringing memory and compute closer together. The company has claimed its architecture could make advanced models faster and cheaper to run, targeting one of the main bottlenecks in large AI systems. Olix, another London-based semiconductor start-up, has raised substantial funding for next-generation AI chips focused on inference workloads. Callosum, backed through state-linked AI investment, is working on hardware-software integration for AI systems.

Britain also has semiconductor activity beyond AI accelerators. Pragmatic Semiconductor has built a manufacturing base in Durham for flexible integrated circuits, targeting low-cost chips used in supply chains, consumer goods and healthcare devices. Its model is different from advanced AI chip fabrication, which remains concentrated in Asia and the US, but it gives the UK a rare manufacturing foothold in a sector where domestic capacity is limited.

The challenge is scale. AI chip development requires large amounts of capital, access to specialist manufacturing, advanced packaging, software ecosystems and customers willing to take technical risk. Nvidia’s lead has been reinforced not only by powerful chips but by its software stack, developer base and deep relationships with cloud providers. UK start-ups may find niches in inference, energy efficiency or novel architectures, but public purchases will need clear performance benchmarks to avoid backing technologies that fail to reach commercial readiness.

The plan is likely to face scrutiny over value for money and procurement discipline. Government attempts to act as a first customer can accelerate promising technology, but they can also expose taxpayers to expensive bets if contracts are poorly structured. Ministers will need to show that purchases are tied to measurable capability, security needs and public-sector use cases rather than symbolic support for national champions.