Inside the UK’s Most Powerful AI Supercomputer
Modular Design Drives Rapid Deployment of Ultra-Efficient Liquid-Cooled System
Earlier this week we shared about the IXP.us combination of modular data centers and AI compute to power university research.
Continuing that theme, here’s a look at an impressive deployment in Bristol, England that leverages modular design to pack extreme density into a compact footprint.
The Isambard-AI system at the Bristol Center for Supercomputing (BriCS) is the UK’s most powerful supercomputer, and one of the most efficient globally.
The 5 megawatt system was delivered in just over a year, and delivers 21 exaflops of AI performance - more computing power than all other U.K. supercomputers combined.
Grace Hopper Superchips Drive Performance
The £225m government-funded national facility was built and run by BriCS in close partnership with HPE and NVIDIA, and is powered by NVIDIA Grace Hopper Superchips.
Some of the system attributes:
21 exaflops of AI performance
5,448 NVIDIA GH200 Grace Hopper Superchips
A Slingshot-11 high-speed interconnect and a 100 GB optical fiber connection
Ranked 11th worldwide on the latest TOP500 list of world’s fastest supercomputers
Ranked fourth globally for energy efficiency
Ground was broken in June 2024. By June 2025, the full 5-megawatt system was live: built, tested and running at a global scale.
Prefabricated modular data halls shipped in and assembled in 48 hours.
Direct liquid-cooling architecture from HPE packs 440 GPUs per cabinet
Overall PUE (Power Usage Effectiveness) below 1.1.
Here’s a closer look at the system in a video shared Thursday by the BriCS team.
The system is named after Isambard Kingdom Brunel, the 19th-century engineer who reshaped Britain with railways, bridges and ships.
It is being used for diverse research projects such as analyzing animal behavior, modeling global financial markets, and projects related to climate change and human health, including vaccine development for conditions like dementia.
On Jan. 22 the Prince of Wales visited BriCS for a tour of the Isambard-AI system, meeting with Professor Simon McIntosh-Smith, Director of BriCS, as well as two researchers using the system’s new capabilities in their work.
Dr. Danielle Paul, Senior Research Fellow at the University of Bristol, whose British Heart Foundation-funded research is using the supercomputer to support the development of new drugs to combat heart disease.
Dann Mitchell, Professor of Climate Science at the University of Bristol, whose research shows how to use AI to better predict extreme weather under different climate scenarios, especially for human and ecosystem health.
Check out our previous coverage of modular systems:
Really impressive deployment timeline. The 48-hour assembly from prefab modules is the real gamechangr here, especially when combined with liquid cooling achieving 440 GPUs per cab. I've watched traditional datacenter builds drag on for years with endless delays and cost overruns. The fact that they hit PUE below 1.1 while packing that density shows how much efficiency you can gain when you optimize the whole stack togther instead of bolting solutions onto legacy architecturee.