The UK AI Chip Plan Bets £400M on Startups Staying Home

The UK artificial intelligence (AI) chip plan unveiled at London Tech Week puts £400 million of public chip purchasing inside a wider £1.1 billion hardware package, giving British AI hardware startups a state-backed first customer while Advanced Micro Devices (AMD, the US chip designer) and Nebius commit a combined £3.7 billion to UK compute capacity.

Prime Minister Keir Starmer’s pitch on 8 June was aimed at the gap between invention and ownership. Britain has chip design groups, photonics startups, AI labs and universities. The new purchase order asks a harder question for companies whose scale customers often sit outside the UK.

A Purchase Order Becomes the Policy Tool

The Department for Science, Innovation and Technology (DSIT, the UK tech department) has wrapped the plan around a new national AI supercomputer. DSIT says the machine will join Isambard-AI, Zenith and Dawn within the AI Research Resource (AIRR, the UK’s public compute programme) and will use a mixed set of processors when deployed in 2030.

Grants remain part of the package. The more unusual piece is an Advance Market Commitment (AMC, a binding promise to buy a product that is still being commercialised). In hardware, that gives a startup a customer to design against before a cloud giant takes the first risk.

• £750 million for a national AI supercomputer that can bring proven and emerging chips into one public system.
• £400 million of that total for next-generation AI chips, split between the early order and a later specialised hardware procurement.
• £150 million set aside this summer for novel inference chips, the hardware used to run AI tools after training.

After the first competition, a further £250 million procurement is meant for specialised hardware as technologies mature. The purchasing route gives DSIT a way to test British-designed chips inside public machines before those companies chase overseas data centre buyers.

Private Compute Money Sets the Scale

The public order sits beside much larger private build-outs announced around the same event. AMD put its number around research and national compute systems. Nebius put its number around local capacity, cloud customers and a London hub.

Nebius, the Amsterdam-headquartered AI cloud company, says London-founded Revolut is already running production workloads on its platform for more than 75 million customers across 40 markets. The demand behind that claim is why capacity stories keep returning to power and site delivery; CoreWeave’s UK data centre lease bet came from the same race for fast access to local AI infrastructure.

The Startup Route Runs Through Testbeds

Small chip firms reach national machines through integration work across boards, racks, software, cooling and benchmarks against incumbent suppliers. The plan gives that work two homes: the new supercomputer procurement and the Scaling Inference Lab run through the Advanced Research and Invention Agency (ARIA, the UK’s high-risk research funder) and CommonAI.

Oriole Networks, a British photonics company, is the named demonstration case. The government says Oriole, working with AMD through the lab, will deploy a large AI system that uses light to move data between chips. AMD describes the work as a test of Oriole’s PRISM photonic networking with its accelerators and server processors for inference workloads.

Fractile and Olix, two British AI hardware startups, have raised more than £320 million between them, according to DSIT. The number sits far below the capital budgets of large US cloud operators. It still puts names to the policy because the state is trying to move domestic suppliers from pilot work into purchase orders.

Capital Is the Second Half of the State Bet

The British Business Bank, the state-owned economic development bank, is working with Playground Global on a UK-focused deep-tech fund backed by a cornerstone commitment of up to £150 million, subject to due diligence and legal negotiations, according to British Business Bank’s Playground Global statement.

The bank says it would be the largest commitment it has made. Playground, a Palo Alto venture firm with more than $1.7 billion under management, plans to focus on university spinouts, research-led ventures and hardware companies that need patient capital before commercial launch.

That target sits beside the £500 million Sovereign AI Unit, a government programme meant to back UK AI companies with funding, compute access and support through Innovate UK and the British Business Bank. Outside government, capital is moving into the same hardware-heavy lane; futurepresent’s $300 million AI infrastructure fund is another sign that investors are treating AI infrastructure as a company-building category with its own capital needs.

Beyond Arm, a Wider Chip Base

Arm, the Cambridge-headquartered chip design company, remains the global name in the UK’s semiconductor story. The Semiconductor Sector Study 2026 commissioned by DSIT counted 705 UK semiconductor companies, including 295 dedicated firms and 408 diversified firms. Dedicated companies generated £10.6 billion in revenue and directly employed about 16,350 people in 2025.

The study also says the sector supports about 26,850 jobs across the economy and remains concentrated in 12 recognised regional clusters. Cambridge, London, Bristol and Southampton carry design strength; South Wales, Scotland and the North East carry more of the manufacturing and materials base.

The study names barriers that appear again in the hardware plan:

• 83% of surveyed firms expect growth over the next three years, while 47% expect annual growth above 20%.
• Talent availability remains a constraint, which is why the plan expands undergraduate bursaries, doctoral training and TechFirst support.
• Scale-up capital and UK operating costs, especially energy, remain barriers as hardware companies move toward production.

The skills line is more than a training annex. The plan lifts total semiconductor and AI hardware skills support to £80 million, including expanded bursaries, a £12 million Centre for Doctoral Training in Chip Design and £20 million of TechFirst support for 500 more UK PhD students.

Starmer Puts Consent Into the Compute Pitch

The policy landed inside a political speech about who benefits from AI. In the Prime Minister’s London Tech Week transcript, Starmer opened with Warrington, where a former Unilever soap factory is being turned into an AI data centre, and later linked AI to faster National Health Service diagnoses, court backlogs and planning decisions.

He also warned technology companies over child safety after the government’s clash with Grok, Elon Musk’s AI chatbot, and called for controls to stop children from sending and receiving explicit images. That safety section sat next to skills numbers: 1.7 million workers have received AI training toward a target of 7.5 million by 2030, and AI tutors are planned for 450,000 children on free school meals.

The chip plan is built for that politics. A supercomputer in Edinburgh, a lab rack for Oriole, a Playground fund and skills bursaries are easier to defend when ministers can point to jobs outside London and compute access for scientists using public machines.

Summer Procurement Will Set the First Price

The first procurement round now carries most of the test. DSIT says the £150 million order will buy novel inference chips this summer, with successful vendors deployed into AIRR and exposed to a range of research workloads. The later £250 million procurement is for additional specialised hardware as designs mature.

That puts a timeline on Starmer’s start, scale and stay line. Founders will know soon whether the state is ready to become the first customer for British AI chips. The first awards are due in summer 2026.