
NVIDIA launched its DSX platform to standardize AI factory design across chips, systems, software and facilities, targeting lowest token cost and faster production. New DSX MaxLPS software uses 45°C liquid cooling to boost GPU performance per megawatt by up to 40%, while DSX OS provides lifecycle management and tenant isolation.
NVIDIA unveiled the DSX platform as a codesigned solution spanning silicon, systems, software and facilities to accelerate AI factory design, deployment and operations. The integrated framework aims to minimize token cost and reduce time to first production by aligning every layer of infrastructure with extreme codesign principles.
DSX MaxLPS leverages 45°C liquid cooling and in-rack optimizations to increase GPU performance per megawatt by up to 40%, delivering lowest token costs for AI workloads. DSX OS brings open source, modular lifecycle management, runtime consistency, health automation and multi-tenant operations, while DSX Sim, Flex and Exchange handle simulation, grid integration and system orchestration.
Cloud providers CoreWeave, Crusoe, Firmus, IREN, Lambda, Nebius, Nscale and Yotta Data Services are deploying DSX components to improve utilization and speed capacity rollouts. OEMs including Dell, HPE, Lenovo, Supermicro, ASUS, Foxconn, GIGABYTE, Pegatron, QCT, Wistron and Wiwynn are building DSX-ready systems, and pilots with Emerald AI and Silicon Valley Power demonstrate grid-responsive AI factories.