UK lines up £250M cloud procurement to feed its growing AI research appetite
Plan would link commercial capacity with Britain's flagship supercomputers
The UK government is looking for cloud providers to support its ambition of increasing its AI compute capacity twentyfold by 2030 in a deal that could be worth £250 million.
The Department for Science, Innovation & Technology announced plans for a procurement that asks for cloud providers to help ensure researchers from academia, public institutions, and SMEs can access the hardware they need.
A pre-market engagement notice said it wanted to "accelerate innovation and deliver transformative solutions in critical domains such as climate science, energy, medicine, and advanced materials, supporting economic growth and public good through Social Value commitments."
The deal is valued at "£214.4 million excluding VAT and £250,000,000 including VAT," according to the notice, which provides information about how DSIT wants to engage with suppliers.
Through the deal, DSIT said it wanted to acquire a solution that integrates with the existing portal run by AI Research Resource (AIRR) – the suite of advanced supercomputers that provides compute capacity to researchers, academia, and industry – which offers access to two existing supercomputers: Isambard-AI at the University of Bristol, and Dawn at the University of Cambridge.
DSIT said it needs a baseline capacity of cloud resources accessible to authorized users via the AIRR Portal, either provided directly or through a brokered, multi-cloud provider model. It is also looking for on-demand scalability to enable access to additional GPU capacity beyond baseline capacity and a managed service that will include secure data storage, orchestration for machine learning workloads, usage monitoring, reporting, demand forecasting and active security management. Technical support and integration is also part of the package.
Billed as Britain's most powerful supercomputer, Isambard-AI came online in the summer. It is based on the HPE Cray EX4000 system and incorporates more than 5,000 Nvidia Grace-Hopper GPUs. It is expected to deliver over 21 exaFLOPS of 8-bit floating point performance for LLM training, and more than 250 petaFLOPS of 64-bit performance.
Meanwhile, Intel, Dell, and the University of Cambridge are developing Dawn, which will also support workloads including academic and industrial research, healthcare, engineering, and climate modeling.
Last week, the government promised a package of reforms and investment to put "AI at heart of government's mission to drive growth, create jobs, and spread prosperity across the country." It includes up to £100 million of government support for British startups.
In the House of Commons last week, minister for digital government and data Ian Murray said that the government's "AI roadmap" had been delayed owing to the change in technology minister in September. ®