PCPro October 2025
We saw the first sign of the UK government’s AI strategy with the recent unveiling of the world’s 11th most powerful supercomputer at the University of Bristol. Isambard-AI is formed of 12 cabinets, each packed with 440 Nvidia Grace Hopper superchips, and is now the cornerstone of AI research in Britain, being free of charge to academic researchers but also open to commercial use – for a price.
“Today we put the most powerful computer system in the country into the hands of British researchers and entrepreneurs,” said the secretary of state for science, innovation and technology, Peter Kyle at the supercomputer’s launch. “Isambard-AI doesn’t just close the gap with our international competitors – it propels the UK to the forefront of AI discovery.”
He then added: “With our AI Research Resource [AIRR] now fully up and running, the UK is home to the raw computational horsepower that will save lives, create jobs and help us reach net zero ambitions faster.”
Beginning of the road
On the same day, Kyle unveiled the UK Compute Roadmap (tinyurl. com/373roadmap). This includes four “purposeful” steps, with the first being to build a “modern public computer ecosystem providing the capacity and certainty UK researchers and innovators need”.
The second is a rather simplistic step of “putting compute to use”, with the third a promise to build a “cutting-edge AI infrastructure so that the UK can play its role at the frontier of AI”. The fourth and final step is to create a “sovereign, secure and sustainable capability”
Across 2025, this means an investment of up to £750 million for a new national supercomputer at the University of Edinburgh. To put this into perspective, Isambard-AI cost around £225 million. The new supercomputer should come online in 2027, and would sit in the top five supercomputers in the world in today’s rankings. The government has also pledged around £250 million to “scale” the UK’s AI Research Resource, and has promised to identify AI Growth Zones across the country – including Scotland and Wales. Northern Ireland is not mentioned in the roadmap. “The UK government’s ambition by 2030 is to have a minimum of 100,000 GPUs available to the market,” said Professor Simon McIntosh-Smith, director of the Bristol Centre for Supercomputing, in a briefing shortly before the supercomputer’s official launch. “If you’re [an] astronomer, you need a telescope. If you’re a data scientist, you need a supercomputer. It’s very simple. You can’t do the work [without it].”
“[The] UK has some of the world’s best data scientists,” added David Hogan, Nvidia’s European vice-president. “It has some of the world’s best AI startups. And the thing that’s missing in that flywheel of making the economy work, from an AI perspective, is compute, and today we have that first element that’s been delivered, but we need to do much more, from a UK perspective, to grow and expand that.”
Where Isambard-AI fits in
There’s little doubt that the UK has fallen behind its rivals when it comes to AI supercomputing. Even with the arrival of Isambard-AI, it sits one place outside the top ten: the USA houses four of the five of the world’s fastest supercomputers, Germany the fourth fastest and Italy has two in the top ten. Aside from Isambard-AI, the UK’s best performers sit in 26th and 63rd place in the official list (last updated in June 2025).
There is a sense of catch-up and hurry-up in the UK’s approach. “Remarkably, it only took 18 months for Isambard-AI to go from being commissioned to built,” said Professor McIntosh-Smith. “Normally this process takes three to five years for such a supercomputer.
“We treated the project like a high-performance processor,” he added. “We executed everything in parallel.” Phase 1 of the project, which included 168 Nvidia GH200 Grace Hopper superchips, finished in summer 2024. And those chips really are super, with each one containing a 72-core CPU and Nvidia’s powerful H100 GPU.
Between January and early June 2025, almost 60 different research projects took advantage, using around 100,000 GPU hours in total. But that’s just the start: Phase 2 saw a huge jump to a total of 5,448 GH200 superchips.
UK academics are already bidding for time on the machine, with one headline project being the “Nightingale AI” model for the NHS. Trained on NHS data, this is a sovereign, multimodal health foundation model. The idea is to personalise care and make diagnoses earlier.
“We are launching a major AI program to harness the vast resources of UK healthcare and biomedical data,” said Professor Aldo Faisal, director of the UKRI Centre for Doctoral Training in AI for Healthcare at Imperial College London, earlier this year when Nightingale AI was first announced. “Our goal is to develop a novel generation of AI foundation models, trained exclusively on healthcare data rather than generic text. This approach will enable us to create a truly transformative system tailored for the medical community.”
Then there’s BritLLM, a British large language model, as you may have guessed. It’s designed to be used by government services – including healthcare and education – and promises to promote inclusivity.
“Isambard-AI is the most exciting development for natural language processing research, and LLM research in particular, in the UK to date,” said Pontus Stenetorp, professor in natural language processing at University College London at ISC 2025.
“We have already been able to produce state-of-the-art, multilingual LLMs with the help of Isambard-AI and over the next few years it is our hope that with access to these resources we will be able to answer fundamental questions about how LLMs operate.”
Isambard-AI is the most exciting development for natural language processing research, and LLM research in particular, in the UK to date
BritLLM is trained on British languages – including Welsh – and aims to promote inclusivity and better delivery of public services such as healthcare and education.
Nvidia’s Hogan believes Isambard-AI offers big opportunities for businesses too. “We also announced a training programme with the government just a few weeks ago,” he said, “where we’re going to train 100,000 people in AI using our deep learning institute programme. So that’s really enabling people to transition their workloads and know how to leverage these AI models.
“And that’s the opportunity. You should see Isambard as almost an incubator for AI. This is a platform: people should come here and build their business on AI, prove that out, and then take that into the world and build another infrastructure and start to scale. This is really the starting point of the journey.”
No comments:
Post a Comment