Summary
- DSIT has opened an expression of interest to identify a host for the AIRR heterogeneous supercomputer; the £750 million programme was announced in June and is not new funding on 17 July.
- Candidate hosts must show a credible path to as much as 10 MW by early 2028 and 20–30 MW in total by FY 2029/30, alongside space, cooling, networking, security and operating capability.
- The eventual host would run the secure and resilient site, while DSIT would retain control of access policy and the allocation of computational time.
Before Britain’s next AI supercomputer becomes a machine, it must become a place. That makes the UK government’s new host-site expression of interest more consequential than a routine request for institutional enthusiasm. It asks where a large block of power can be secured, which organisation can operate national research infrastructure, and how public control will be divided once the hardware arrives.
The chronology matters. The government announced the £750 million heterogeneous AI supercomputer in June as part of its AI Hardware Plan. What changed on 17 July was the opening of a process to test the interest and ability of potential hosts. DSIT explicitly says the exercise is not a call for competition, an invitation to tender or a commitment to award funding or a contract. The new fact is therefore a selection-stage move, not another spending announcement.
Location becomes the first systems test
The proposed system is designed around a spine of established computing hardware, supplemented by inference-specialised modules, advanced storage and networking, and a software layer capable of coordinating heterogeneous workloads. The plan also leaves room for a future quantum module. That mix makes physical hosting an integration problem rather than a real-estate choice.
DSIT expects a phased deployment. Phase 1 is intended to go live in early 2028 and could require up to 10 MW. Phase 2, planned for FY 2029/30, would bring total demand to 20–30 MW, including the first phase. Respondents must describe existing and prospective power capacity, what still needs consent or construction, backup arrangements, cooling, networking, security and the evidence behind their delivery dates.
Those questions expose the binding constraints on sovereign compute. Accelerators can be procured in stages; grid connections, high-density cooling and secure buildings cannot be improvised on the same timetable. A candidate with strong research credentials but an uncertain power path may be less deliverable than an institution that can show ready space, credible expansion and high-bandwidth access to national research networks.
The host is an institution, not merely a landlord
Eligibility narrows the field. A bid can come from one organisation or a partnership, but a consortium must be led by a UK-based research organisation eligible for UKRI funding. The lead must have UK legal personality and demonstrable experience of operating high-value, large-scale compute services for research or innovation users at national level.
The EOI also asks for senior institutional backing, governance arrangements, operating and user-support models, relevant accreditations and experience integrating novel hardware. Security expectations include ISO 27001 and Cyber Essentials Plus, or a credible route to obtain them by December 2028, together with the ability to meet an enhanced Cyber Assessment Framework profile for supercomputing.
That institutional test defines the control surface. The selected host would be responsible for operating the site and maintaining the service’s security and resilience. DSIT, however, would retain the access policy and decide how computational time is allocated under the AIRR programme. Sovereignty here is not simple ownership of processors. It is a division of authority across the department that controls access, the institution that runs the service, and the suppliers that provide hardware and supporting infrastructure.
The economics are still deliberately open
The £750 million system budget is already divided in broad terms. DSIT says £150 million is intended for an Advanced Market Commitment for novel inference capability, a further £250 million for specialised hardware supporting the second phase, and the remaining £350 million for established-vendor hardware plus data-centre and associated infrastructure costs. Additional operating expenditure or money for a quantum module may follow.
None of that makes the EOI a £350 million hosting award. Funding arrangements for data-centre infrastructure are still being developed and would be agreed with any selected host. Further exercises are expected for system procurement, service provision and other eligible costs.
Respondents are nevertheless being asked to reveal the cost stack now: land, site preparation, power connections, security, networking, construction, electrical equipment, cooling, professional services, staff, user support and annual energy exposure. That makes the exercise an early price-discovery mechanism as well as a capability test. It should help DSIT judge whether the stated budget can buy not only hardware but also a durable operating environment.
What the next decision will reveal
Expressions of interest are due by 11:59pm BST on 10 August. DSIT may then seek clarifications and invite a shortlist into a future selection process; the Secretary of State would make the final host decision. Until that happens, there is no selected site, supplier or construction contract.
The important signals will be less about institutional name recognition than about evidence. A credible response must align power, site readiness, security certification, specialist staffing and national user support against the same timetable. It must also show how the investment would create reusable public assets and skills, strengthen resilience, and fit the regional distribution of UK research capability.
This is why the EOI matters. The British sovereign-compute programme has moved from an ambition expressed in pounds to a contest over place and operating authority. Its first hard constraint may not be the choice of accelerator. It may be the ability of one research-led institution to assemble enough electricity, infrastructure and governance to run the system as a secure public service.