Isambard-AI combines highly advanced systems architecture and rapid deployment with world-leading sovereign AI stack to deliver the UK’s most powerful AI computing capability.
Supercomputers are the foundation of modern research and innovation. They power ground-breaking projects in areas critical to our lives such as climate modelling, drug discovery, LLM training and AI security, enabling researchers to process vast datasets and simulate complex systems at unprecedented pace.
Isambard-AI is the UK’s most powerful supercomputer. In November 2025, it ranked 11th most powerful in the world on the TOP500, the international league table of private and public compute.
Delivered through the Bristol Centre for Supercomputing (BriCS) at the University of Bristol, Isambard-AI forms a central pillar of the UK’s AI infrastructure ecosystem. It is a core component of the AI Research Resource (AIRR), funded by the Department for Science, Innovation and Technology (DSIT) and provides the compute for Sovereign AI, funded by UK Government.
Following its launch in July 2025, access to Isambard-AI opened to academia and industry via AIRR; with the launch of the Sovereign AI fund, this further extended to start-ups and innovators. This ensures that the benefit of the supercomputer can be applied wherever it will have the biggest impact on the UK economy.
A system built for AI
Isambard-AI has been designed specifically for AI workloads, particularly large-scale model training, AI inference, and data-intensive simulation.
The system is built on the HPE Cray EX platform, delivered in partnership with Hewlett Packard Enterprise. At its core are 5,448 NVIDIA GH200 Grace Hopper superchips, each tightly coupling a CPU and GPU into a single architecture optimised for AI.
These superchips are organised into 1,362 compute nodes, with four superchips per node. Each node operates with a unified memory space of around 864 GB, which is critical for handling large AI models efficiently without excessive data movement.
Across the full system, the hardware delivers over 21 exaflops of AI performance, enabling large-scale model training at global-leading scale; around 200 petaflops of traditional HPC performance, supporting scientific workloads and nearly 25 petabytes of high-performance storage, designed for data-intensive AI pipelines.
The system is connected using HPE Slingshot interconnect technology, providing high-bandwidth, low-latency communication across nodes.
Energy and sustainability as considerations
Data centres have a reputation for using lots of electricity, powering both the hardware and cooling systems. BriCS tackled this problem head-on with the design of Isambard-AI.
First, all the power used by Isambard-AI comes from entirely renewable sources, in the UK.
Isambard-AI uses direct liquid cooling (DLC), removing the need for fan-based systems and significantly reducing energy overhead.
Second, Isambard-AI uses the most energy efficient technology to keep the system cool: direct liquid cooling (DLC). This removes the need for energy inefficient fan-based systems. Instead, Isambard-AI’s DLC system sends a mix of coolant and water around a closed loop, which runs through small pipes in the computing hardware itself, known as blades. Heat is transferred from the blades to the liquid which is then taken away to be cooled outside, before being delivered back into the system.
In addition, the system itself is housed in a modular data centre (MDC). BriCS chose this hugely innovative approach in line with HPE’s MODPOD design, which enabled rapid deployment while reducing construction emissions. There’s no building to speak of, no toilets or offices. Isambard-AI sits inside what looks like four large shipping containers, attached together, protected by rigorous security defences.
This approach resulted in:
- A build that cut construction emissions by approximately 72% compared to traditional data centres
- A class-leading power usage effectiveness (PUE) of around 1.06, meaning only 6% of the total energy used to power the computer is used additionally to power the cooling system. Traditionally, air-cooled supercomputers have PUEs of around 1.9 to 2.0, meaning they used 90% to 100% additional power for the cooling system.
The Isambard-AI project has proven that, when building AI infrastructure, performance and sustainability can be engineered together. As a result of all of this design and thought, Isambard-AI launched at #2 on the Green500 list of supercomputer efficiency.
From funding to full system in under two years
Traditional supercomputers of this scale typically take three to five years from funding to deployment, whereas Isambard-AI moved at unprecedented speed, going from concept to operation in under two years, with the build phase delivered in under 18 months.
The timeline was unusually compressed:
- Late 2023 – Government investment confirmed through DSIT and UKRI under AIRR
- Early 2024 – System design finalised, phase 1 system procured and installed
- Mid 2024 – Main site construction
- Late 2024 – Early users access the phase 1 system for pilot workloads, phase 1 modular data centre arrives
- Early 2025 – Phase 2 system installed
- Mid 2025 – Full system operational, acceptance passed, real users from August
This pace reflects a deliberate shift in how national infrastructure is delivered, moving closer to the speed of industry than traditional public-sector programmes.
AIRR and Sovereign AI
Through the AI Research Resource (AIRR), Isambard-AI is being positioned as a national resource for all. Access is allocated through structured calls, prioritising projects that deliver clear scientific, economic, or societal impact.
AIRR enables broader access across academia, industry, and the public sector; allocation is based on merit and impact, and alignment with national priorities, from healthcare to climate science.
Through Sovereign AI, funded with £500m from UK Government, portfolio companies are provided with dedicated access to Isambard-AI. This addresses some of the most significant constraints in AI development, the cost of compute and the infrastructure required to innovate.
Training advanced models requires access to supercomputers optimised for AI workloads. Without infrastructure at this scale, many companies simply cannot compete, and many start-ups go abroad to scale.
By integrating compute access into funding packages, the Sovereign AI programme is effectively lowering the barrier to entry for frontier AI development, linking public investment directly to commercial innovation.
Successful Sovereign AI applicants gain access not only to the Isambard-AI system itself, but also to expertise from the Bristol Centre for Supercomputing, helping them optimise workloads and scale effectively.
Applications at scale
Isambard-AI is helping British Heart Foundation researchers to model proteins in the quest to identify treatments for cardiomyopathy; training UK LLMs and including minority ethnic UK languages such as Welsh, Gaelic and Cornish; identifying skin tone bias in skin cancer detecting apps; understanding how to best train AI agents to ensure they are secure and how easy they are to corrupt; modelling pollution to within 1sqm of the whole of the UK and tracking cattle who may be displaying signs of sickness to support agricultural efforts.
This is a UK national treasure which will benefit the whole world.