NVIDIA Launches DSX Ready to Standardize Power and Liquid Cooling in AI Factories

NVIDIA announced the launch of NVIDIA DSX Ready, a formal qualification program designed to certify power and thermal infrastructure products against NVIDIA DSX AI factory reference designs. The program launches with two primary hardware categories: battery energy storage systems (BESS) from partners including Tesla, Hitachi Energy, and LG Energy Solution, alongside cooling distribution units (CDUs) from Vertiv, LiquidStack, and LG Electronics. To achieve certification, vendors must execute standardized test suites and submit verification telemetry ensuring seamless interoperability with high-density rack architectures. This development directly targets the severe facility-level constraints governing modern AI infrastructure deployment. For platform engineers and datacenter operators, scaling accelerator clusters to hundreds of thousands of GPUs is no longer just a compute or networking problem; it is an electrical and thermodynamic engineering challenge. Uncoordinated power draw spikes during multi-node collective operations can destabilize local power grids, while inadequate liquid cooling flow rates cause localized thermal throttling across dense accelerator nodes. By certifying BESS and CDU hardware to standardized operational boundaries, the program reduces procurement risk and eliminates custom one-off facility adaptations. Within the broader cloud and AI systems ecosystem, the boundary between chip design and datacenter facility engineering is rapidly dissolving. Hyperscale infrastructure has shifted from generic rack deployments to tightly co-designed AI factories where power delivery, liquid cooling manifolds, high-speed optical fabrics, and silicon telemetry operate as a single coordinated system. As rack power densities exceed 100kW to support architectures like Vera Rubin NVL72, traditional facility topologies fail. Programs like DSX Ready, paired with broader initiatives like the AI Energy Management Alliance, formalize this shift by integrating grid-responsive power controls and direct-to-chip liquid cooling directly into the AI hardware baseline. In practice, infrastructure engineering teams planning new AI datacenter buildouts or high-density colocation expansions should incorporate DSX Ready certifications into their vendor evaluation criteria and RFP processes. Utilizing qualified cooling distribution units minimizes the risk of fluid pressure drops and thermal decoupling during sustained peak compute phases. Furthermore, integrating qualified battery energy storage systems enables facility architects to safely implement peak-shaving and dynamic load-shifting strategies, maximizing useful compute capacity under strict megawatt-capped utility interconnect agreements without risking hardware outages.