UK compute access gap threatens startup ambitions

Britain’s AI ambitions face an infrastructure bottleneck: smaller companies struggle to secure reliable, affordable computing capacity while major providers prioritize customers able to make large, long-term commitments. An Oxford University study found that just 32 countries host specialized AI data centers, with US and Chinese firms operating more than 90% of them. That concentration leaves startups and entire nations dependent on infrastructure controlled elsewhere.

Demand for accessible compute is already substantial. The government’s Sovereign AI Fund received over 200 applications requesting more than 67 million GPU hours in just three months. Yet shortages coexist with underused hardware. An analysis of around 23,000 cloud clusters found average GPU utilization of 5%, reflecting companies reserving capacity for peak demand and retaining resources they do not consistently use.

A proposed response is to pool GPUs across universities, businesses, regional data centers and specialist providers into a shared marketplace under UK jurisdiction. Public funding could purchase access for startups, complementing investment in national supercomputers rather than waiting for new facilities. Procurement policies and easier movement of workloads between providers could also reduce dependence on individual clouds, encourage competition and spread infrastructure benefits beyond London and the South East.