Technology

UK government seeks hosts for new £750 million heterogeneous AI supercomputer

The Department for Science, Innovation and Technology has invited UK research organisations and consortia to bid for hosting a major new national AI system designed to bolster sovereign capabilities in frontier research and secure compute.
Listen
AI-generated image: UK government seeks hosts for new £750 million heterogeneous AI supercomputer
AI-generated image for illustrative purposes.
Intelligent summary
  • The Department for Science, Innovation and Technology issued an expression of interest on 17 July 2026 for hosts of a £750 million heterogeneous AI supercomputer within the AI Research Resource.
  • The system combines established and novel hardware to enable frontier AI research, large-scale inference and scientific discovery, with phase one targeted for early 2028 at up to 10 megawatts.
  • Eligible UK research organisations and consortia must demonstrate capability to deliver a secure national compute service, building on existing facilities such as Isambard-AI, Dawn and Zenith.

In a nondescript government building in London last Friday, officials quietly issued a call that could shape Britain's place in the global race for advanced computing power. The Department for Science, Innovation and Technology published an expression of interest seeking hosts for a new heterogeneous AI supercomputer, backed by a £750 million investment as part of the AI Research Resource.

This move stands out for its emphasis on keeping critical infrastructure under domestic control. Rather than depending on distant cloud providers or foreign hardware giants, the initiative prioritises UK-based organisations with the expertise to run a secure, resilient national service. Eligible hosts, whether individual research bodies or collaborative consortia, must show they can deliver reliable access for researchers, businesses and public sector users alike.

The system itself will blend established hardware with newer, experimental designs. Such heterogeneity is intended to support not only cutting-edge AI model training but also large-scale inference tasks and broader scientific discovery. Phase one deployment is eyed for early 2028, drawing up to 10 megawatts of power, with a second phase in 2029-2030 scaling to between 20 and 30 megawatts. Those figures hint at the sheer physical demands of hosting what officials hope will rank among the world's most advanced systems by the end of the decade.

Building on existing foundations

The AI Research Resource already operates several notable facilities. Isambard-AI at the University of Bristol, Dawn at the University of Cambridge and Zenith, also in Cambridge, form the current backbone. The new supercomputer is positioned to expand this network substantially, creating a layered national capability that integrates potential future technologies such as quantum elements.

Hosts will shoulder significant responsibilities: site operations, physical security, and the engineering resilience needed to keep the system running without interruption. Yet ultimate oversight of who gains access remains with the department. This division keeps strategic direction in public hands while drawing on institutional know-how scattered across the country's research landscape.

The expression of interest sets a brisk timeline. A briefing session is scheduled for 27 July, with formal responses due by 10 August. No host has been chosen yet. The assessment process will culminate in a decision by the Secretary of State, ensuring the final selection aligns with national priorities rather than local boosterism.