Liquid Cooled AI Workload Case Study at VIRTUS

This case study details the implementation of a liquid cooled AI workload at VIRTUS Data Centres, a leading provider of colocation data centres in London. The project was initiated by an existing customer, referred to as Customer A, who required a multi-megawatt AI workload with a power density exceeding the capabilities of traditional air-cooling solutions. The challenges included managing high kW density per rack, cooling design responsive to fluctuating compute demands, and meeting stringent environmental, social, and governance (ESG) requirements. VIRTUS developed a tailored solution that incorporated both standard cooling for networking IT servers and a direct-to-chip liquid cooling system for the AI deployment. The design ensured structural integrity to support the weight of the equipment and utilized 100% renewable energy sources to minimize environmental impact. The successful commissioning demonstrated the system's ability to operate efficiently under varying loads, achieving a low operational Power Usage Effectiveness (PUE) of less than 1.1. This project provided valuable insights for future data centre designs accommodating AI workloads.