NVIDIA Unveils DSX Data Center Platform Focused on Full-Stack AI Factory Efficiency

Key summary

NVIDIA has introduced a new data center platform called DSX, distinct from its existing DGX and HGX offerings. While DGX utilizes NVIDIA components and HGX supports mixed CPU or chip architectures, DSX is designed as a full-stack AI factory reference architecture for the entire data center rather than individual servers. The platform includes software tools for dynamic energy allocation (DSX MaxLPS), grid-aware workload management (DSX Flex), simulation testing (DSX Sim), infrastructure configuration (DSX OS), and server-to-management system connectivity (DSX Exchange). According to NVIDIA, the DSX Reference Design supports closed-loop liquid cooling with air-based heat rejection, aiming for near-zero water usage. The company claims this design could save approximately 54 million liters of water daily for a gigawatt-scale data center. Energy management software is projected to reduce energy consumption by 30%, while grid-aware solutions in the UK tests showed a 40% reduction in grid power draw by utilizing renewables or batteries. NVIDIA has partnered with industrial and energy firms including AVEVA, Dassault Systèmes, Siemens, ABB, Eaton, GE Vernova, Hitachi Energy, Schneider Electric, Siemens Energy, Trane Technologies, and Vertiv to test solutions alongside DSX. Several cloud providers and data center operators, such as CoreWeave, Firmus, Google Cloud, IREN, Lambda, Nebius, Nscale, and Sharon AI, have announced support for the platform, with adoption planned for the era of the Vera Rubin chip.