On a damp and humid Thursday morning on the outskirts of Bristol, I join a small gathering of academics, enterprise execs and media for the official ribbon-cutting of Isambard-AI—the UK’s most powerful supercomputer.

The mood is quietly expectant. While the drizzle dampens the grass outside the Bristol and Bath Science Park, inside there’s a growing sense that this newly unveiled system marks a fundamental shift in how Britain approaches artificial intelligence.

Developed by Hewlett Packard Enterprise (HPE) and backed by £225 million in government funding, Isambard-AI is not just another data centre installation—it’s considered a national asset, with the Secretary of State for science and innovation Peter Kyle due to officially announce its opening in the afternoon.

The supercomputer, which combines more than 5,400 Nvidia Grace Hopper superchips, can complete in a single second what the entire global population would take 80 years to calculate manually.

With its peak performance of 21 AI exaflops (that’s 21 quintillion AI calculations per second), Isambard-AI joins the top tier of Europe’s AI infrastructure—and brings home capabilities that, until now, UK researchers and businesses had to seek abroad.

“While large language models and generative AI are grabbing headlines, the compute behind them has largely sat overseas,” says Simon McIntosh-Smith, director of the Bristol Centre for Supercomputing. “With Isambard, we can now train those kinds of models here in the UK.”

Isambard

Simon McIntosh-Smith, director of the Bristol Centre for Supercomputing

 

The implications are broad—and already visible in projects ranging from personalised drug discovery to dairy herd monitoring.

Domain-specific models at scale

 

One of Isambard’s most significant contributions is its ability to power sovereign, sector-specific AI models—those trained not on vast swathes of internet data but on curated datasets unique to sectors such as healthcare, agriculture, or legal services.

This means UK organisations can build systems attuned to local regulatory frameworks, industry language, and public service needs.

In life sciences, that capability is already being tested. Dr Jon Lees and his team are using the supercomputer to map how proteins interact within human cells—an essential step in drug development.

“We’ve been using newer versions of AlphaFold to study around 200 key human proteins,” he says. “These include those related to ageing, Alzheimer’s and various cancers.”

The difference Isambard makes is less about theory than practicality. “On my own GPU machine at home, reconstructing all the cell interactions would take 30 to 50 years,” Lees says. “With a day on Isambard, I can do it.”

One promising target is a cardiac protein associated with hypertrophic cardiomyopathy, a common genetic heart condition. “We found a strong hit for it in the genome—something previously unknown. If we validate it experimentally, it could be the basis for a new therapeutic.”

A coffee with…Jelle Prins, Cradle.bio

Once these structures are mapped, the team can begin rational drug design—using AI to generate small molecules or antibodies that can bind to the proteins. The more simulations they can run, the higher their chances of finding promising candidates. “Scale is everything,” says Lees. “The more we scale, the better hits we can generate.”

Down at the farm

 

A few miles away, the supercomputer is solving very different problems—but with similarly tangible results. At the University of Bristol’s John Oldacre Centre for Dairy Welfare & Sustainability Research, researchers are using AI to monitor cow health in real time.

With over 60 cameras installed across the facility, Isambard-AI is training models to detect early signs of illness based on subtle behavioural cues—such as posture changes or social withdrawal.

By intervening early, farmers could boost productivity and animal welfare, while also better managing issues like antimicrobial resistance and greenhouse gas emissions.

In another project, researchers from the University of Southampton and the Rosalind Franklin Institute are exploring the intricacies of placental biology. Their aim: to understand how the organ’s structure influences pregnancy outcomes and lifelong health.

“It’s not just about looking at pretty images,” says Dr Laura Shemilt of the Rosalind Franklin Institute. “We want to model the placenta as a complete, integrated system.”

Isambard AI

University of Bristol researchers are using Isambard AI to monitor cow health in real time

 

Traditionally, this work requires laborious manual annotation—scientists tracing structures pixel by pixel. With Isambard-AI, that bottleneck could be broken.

“We’re training AI to segment high-resolution 3D placenta images,” she explains. “That lets us extract far more quantitative data—and start linking structure to function.”

Shemilt says the team has used around 30% of its allocated 13,000 hours of compute time, with much of the effort focused on scaling their models using the Grace Hopper architecture. “The segmentation problem is hard—especially in 3D. But it’s one we think this system can crack.”

A platform for industry

 

Though based at a university, Isambard-AI isn’t designed for academics alone. More than 80 teams have already applied to use the system, including start-ups and SMEs looking to train models or run intensive AI workflows.

“There aren’t many dedicated AI research supercomputers like this,” says McIntosh-Smith. “In our first open call, which only ran for a few weeks, we had enough demand to fill the machine for a year.”

Access is currently allocated by the UK government, though the system is designed to be enterprise friendly.

Isambard: UK Tech and Science Secretary Peter Kyle unveils UK's new 'national treasure'

UK Tech and Science Secretary Peter Kyle unveils the nation’s new ‘national treasure’

 

“We’ve built it so companies used to running code on public cloud can come in, drop their containers, and start training models almost immediately,” he explains.

He adds that much of the ecosystem uses open-source software, and cybersecurity measures—including penetration testing and a full cyber assessment—ensure commercial-grade safety.

The cost model, still evolving, is intended to match or undercut public cloud providers without undermining the commercial sector. “We’re pricing it so that it’s affordable, but not so cheap that it distorts the market,” McIntosh-Smith adds.

For Nvidia’s VP of enterprise EMEA, David Hogan, the value lies in how organisations use the system as a proving ground. “You should see Isambard as almost an incubator for AI. People can build their business here, prove the models, and then scale up elsewhere.”

Building a sovereign AI future

 

The strategic importance of developing AI capacity within UK borders is a recurring theme for those behind the project. “The sovereign AI situation is real,” says McIntosh-Smith. “We can now train something like GPT from scratch in the UK. A latest-generation model would take us a few weeks.”

A fraction of the system—around 5%—is reserved for University of Bristol research. The rest is being opened in waves to academic and enterprise users across the country. From life sciences and language modelling to legal tech and climate analytics, the range of applications is only widening.

Researchers working on BritLLM—a UK-focused large language model—can now train full-scale models rather than downscaled versions limited by compute constraints. And for academics like Dr Lees, the system represents not just power, but opportunity.

 “What Isambard gives us is democratisation,” he says. “This work would have cost us a fortune. Now we can try things out, generate preliminary data, and build the case for future funding.”

There are environmental benefits too. Ranked the world’s fourth most energy-efficient supercomputer, Isambard-AI is cooled by water, not fans. The resulting waste heat is piped into the adjacent University of Bristol National Composites Centre—and could eventually help warm nearby homes and businesses in South Gloucestershire.

“We’ve already put in the plumbing,” says McIntosh-Smith. “We’re serious about using this infrastructure as efficiently as possible—not just in terms of energy, but what it unlocks.”

What’s next?

 

With 25 petabytes of flash storage and internet-based access for users across the UK, Isambard-AI is designed to scale. More enterprise-focused initiatives are expected in the coming months, along with deeper partnerships in the NHS, legal sector, and UK manufacturing.

Above all, its launch shifts the balance in UK AI development—from relying on overseas platforms to creating the infrastructure for domestic innovation. “This isn’t just a machine,” says McIntosh-Smith. “It’s a catalyst for everything that follows.”

Personalized Feed
Personalized Feed