#Industry News
NVIDIA Liquid Cooling Supply Chain, the Component System and the Sealing Technology
362kW Fanless Vera Rubin, L10 Assembly and Four A-Group Plate Suppliers: How NVIDIA's Long-Arm Management Reshapes the Liquid Chain
AI compute demand is rewriting data center cooling and the NVIDIA chain with it: AI server output passed 400 billion dollars in 2025 with rack power beyond 120 kilowatts, Vera Rubin goes all-liquid and fanless at 362 kilowatts per rack with per-unit module demand growing from two to seven, and long-arm management shifts assembly from the L6 phase to the final L10 phase exclusive to Foxconn, Wistron and Quanta while cold plate purchasing centralizes around four A-group suppliers led by AVC and Cooler Master. Quick couplings carry the highest barrier with JONHON, Staubli and CPC among the few certified firms, JONHON holding about 90 percent of the domestic UQD market, while the inner manifold carries ten to thirty weld points per unit and the CDU loop adds more than twenty potential leak points under the 800-volt DC architecture. EXTRCOOL provides one-stop liquid cooling solutions.
