DCX Liquid Cooling Systems unveiled an 8MW warm-water cooling solution for Nvidia's next-generation Vera Rubin AI chips.

DCX Liquid Cooling Systems announced the second generation Facility Distribution Unit (FDU V2AT2), a new CDU system designed to support the infrastructure shift driven by AI deployments at scale. The system delivers up to 8.15MW of heat transfer capacity with record-breaking flowrate, and specifically supports 45°C warm-water cooling for NVIDIA's NVL72 GB200 GB300 Blackwell and Vera Rubin architectures. The FDU V2AT2 represents the most powerful CDU system ever made by the company, featuring what DCX describes as the most capable heat exchanger in the industry.

According to Maciek Szadkowski, CTO at DCX, "As the datacenter industry transitions to AI factories, operators need cooling system that won't be obsolete in one platform cycle. The FDU V2AT2 replaces multiple legacy 1.3MW CDUs and enables 45°C supply water operation. This new category of CDUs has minimum thermal loss with AT2 approach temperature, and provides multi-megawatt cooling at the hall level. That opens a clear path to NVIDIA Vera Rubin architecture and beyond while simplifying cooling loop topology and significantly reducing both CAPEX and OPEX of datacenter liquid cooling system."

The 45°C warm-water capability enables a significant operational shift in data center cooling. NVIDIA has highlighted 45°C water supply as a key design point for Vera Rubin rack-scale systems, explicitly enabling data centers to operate without traditional water chillers in the cooling plant for these workloads. The FDU V2AT2 supports ASHRAE W45 W+ warm-water classes and maintains tight control to avoid condensation, helping operators transition from refrigeration-centric cooling to a simpler, more scalable warm-water approach while enabling heat reuse and chillerless heat transfer.

Unlike CDUs designed for a single row or small cluster, the FDU V2AT2 is engineered for facility-scale cooling topology that supports complete data hall installations, efficiently serving large-scale, high-density AI server racks with high flow and high availability. The system is available for hyperscale and high-density AI data center projects effective immediately, with reference architectures, integration guidance, and project sizing support available upon request.