The physical limits of artificial intelligence are no longer an abstract concern for universities. As models grow larger and simulations more complex, the question facing research institutions is not simply how much compute they can access, but whether the infrastructure supporting it can scale sustainably. A new deployment at a UK data centre suggests that liquid cooling is moving from niche experiment to core requirement.
Digital Realty has confirmed that Imperial College London has selected its Woking facility to host the university’s first direct liquid-cooled high performance computing environment in the UK. The system represents Imperial’s most advanced AI and HPC deployment to date and marks a shift in how academic institutions are approaching the infrastructure needed to support data-intensive research.
The installation is designed to underpin work across artificial intelligence, climate science, healthcare, engineering and fundamental research, while reinforcing Imperial’s position as one of the UK’s largest Tier 3 university HPC environments. More significantly, it reflects a growing recognition that AI research is now constrained as much by power density and energy efficiency as by access to algorithms or data.
From ambition to infrastructure reality
The Woking deployment builds on Imperial’s ICICLE initiative, first announced several years ago with the aim of developing advanced, energy-efficient AI and HPC capabilities through collaboration with technology partners. Moving into 2026, that ambition has translated into a production-scale system housed in a commercial, AI-ready data centre rather than on campus.
The platform uses bespoke direct liquid-cooling technology supplied by Lenovo, enabling far higher compute density per rack than traditional air-cooled systems. Liquid cooling allows heat to be removed directly from processors, supporting more powerful workloads while reducing overall energy consumption, an increasingly important consideration as universities face both sustainability targets and rising electricity costs.
For Imperial’s researchers, the practical impact is clear. The system enables the training of larger and more complex AI models, accelerates data-heavy simulations and shortens the path from research insight to real-world application. Fields such as drug discovery, robotics, climate modelling and advanced materials all depend on the ability to iterate rapidly at scale, something that conventional HPC environments are struggling to deliver.
Data centres as national research infrastructure
The project was delivered on an accelerated timeline. The contract was signed in July 2025, with the fully operational environment handed over in the final quarter of the year. That process included structural floor reinforcement, installation of the liquid-cooling infrastructure, system integration and commissioning, underscoring the level of engineering required to support modern AI workloads.
This matters beyond a single university. As AI models become more compute-intensive, national research strategies increasingly depend on access to specialist data centre infrastructure capable of supporting high-density, high-efficiency systems. Traditional campus facilities are often ill-suited to this task, lacking the power delivery, cooling capacity and resilience required.
By locating the deployment within a purpose-built data centre, Imperial gains access to an environment designed for extreme performance and availability, while avoiding the need to retrofit ageing buildings. For Digital Realty, the project demonstrates how commercial operators are becoming critical partners in the UK’s research ecosystem, not just hosting data but enabling the next generation of scientific discovery.
The collaboration also highlights a broader shift in how AI infrastructure is conceived. Liquid cooling is no longer positioned as an experimental option for edge cases, but as a foundational technology for AI and HPC systems where performance and sustainability must advance together.
Taken together, Imperial’s research leadership, Digital Realty’s data centre platform and Lenovo’s liquid-cooled computing technology offer a blueprint for how the UK can scale AI capability responsibly. As demand for compute continues to rise, the lesson is increasingly clear. The future of AI research will be shaped as much by cooling pipes and power density as by code.
