NVIDIA offers DSX AI factory playbook

NVIDIA said on May 31 that it launched the DSX platform, a packaged set of reference designs, software, APIs and deployment guidance for companies building what it calls AI factories. The company announced the product at GTC Taipei and described it as a “complete playbook” for designing, simulating, building and operating AI infrastructure at scale. In NVIDIA’s materials, DSX is presented as a common architecture spanning chips, systems, facilities software and partner technologies. The company said the goal is lower token cost, faster deployment and higher operational resiliency. ### What is NVIDIA actually selling with DSX? NVIDIA said DSX combines open-source modular software libraries, APIs, reference designs, NVIDIA compute platforms and partner technologies into one codesigned platform. The company’s documentation breaks that into several pieces: DSX Sim for simulation and digital twins, DSX OS for operational software, DSX MaxLPS for power and performance management, DSX Flex for power orchestration, DSX Exchange for IT and OT data exchange, and hardware and facilities reference designs. (nvidianews.nvidia.com) Jensen Huang, NVIDIA’s chief executive, framed the offer as more than a component list. “We’re not just shipping chips — we’re giving every infrastructure builder a complete playbook to build AI factories,” Huang said in the company’s announcement. He added that customers could simulate the factory before spending on hardware and validate performance before racks are installed. (nvidianews.nvidia.com) ### Why does the company keep using the term “AI factory”? NVIDIA’s product pages say DSX is designed around “lowest token cost” and “tokens per watt,” language that treats AI infrastructure as an industrial production system rather than a conventional data center. The company says DSX AI factories are “co-designed as unified products” across compute, power, cooling and software, with the stated aim of improving bring-up time, resiliency and output efficiency. (nvidianews.nvidia.com) Warren Barkley, in an NVIDIA technical blog published May 31, wrote that AI factories must coordinate energy, chips, infrastructure, models and applications across a five-layer stack. In that post, NVIDIA said power is the limiting factor in an AI factory and described DSX as connecting grid behavior, facilities controls and AI infrastructure rather than treating them as separate systems. (nvidia.com) ### What are the most concrete technical claims? NVIDIA said DSX MaxLPS combines 45-degree-Celsius liquid cooling with in-rack technologies to optimize performance per watt. In the press release, the company said that setup lets operators run up to 40% more GPUs at their most energy-efficient operating point with minimal impact on workload performance. (developer.nvidia.com) The documentation and product pages also emphasize simulation before deployment. NVIDIA says DSX Sim includes high-fidelity logical simulation, validated integrations and the Omniverse DSX Blueprint for digital twins of gigawatt-scale AI factories. The company says those tools are intended to model, validate and optimize infrastructure decisions before and after physical deployment. ### How far does this push NVIDIA beyond chips? (nvidianews.nvidia.com) NVIDIA said DSX OS is open-source and modular software for lifecycle management, scheduling, runtime consistency, health automation, resiliency and multi-tenant operations. In its technical blog, the company said some of that software is drawn from infrastructure and platform software it already uses on DGX Cloud and is now releasing as open source for partners. (docs.nvidia.com) That systems role is already showing up in customer and partner announcements. On May 7, NVIDIA and IREN said they planned to support deployment of up to 5 gigawatts of DSX-aligned AI infrastructure across IREN’s global data-center pipeline over time, with future deployments expected to focus on IREN’s 2-gigawatt Sweetwater campus in Texas. (nvidianews.nvidia.com) ### Where would a builder start if it wanted to use DSX? NVIDIA’s DSX documentation site now serves as the implementation map. The company’s product page lists the reference designs, simulation stack, operations software, power-management tools and exchange layer as the main building blocks, while linked resources point developers to Omniverse DSX Blueprint materials and DSX OS components such as Run:ai, Kubernetes-native scheduling and cloud functions. (nvidianews.nvidia.com) May 31 is the launch date NVIDIA gives for the platform, and the next public artifacts are already live in its documentation, product pages and technical blogs. NVIDIA has also tied DSX to ongoing deployments, including the IREN partnership announced on May 7 and OEM ecosystem support described in its GTC Taipei materials. (nvidianews.nvidia.com) (docs.nvidia.com)