Defining the NVIDIA DSX Platform for AI Factories
NVIDIA DSX is a full-stack AI factory software platform that unifies chips, systems, facilities, and partner technologies into a common architecture for designing, simulating, building, and operating large-scale AI infrastructure as standardized “factories” for generating tokens. Rather than treating compute, networking, storage, power, and cooling as separate projects, the NVIDIA DSX platform gives infrastructure builders an integrated framework that runs from initial design through deployment and ongoing operations. It combines open source, modular software libraries, APIs, reference designs, and NVIDIA accelerated computing systems into one coordinated stack aimed at lowering token cost and shortening time to first production. As NVIDIA CEO Jensen Huang puts it, “We’re not just shipping chips — we’re giving every infrastructure builder a complete playbook to build AI factories,” highlighting DSX as both technology and methodology for AI infrastructure management.
Inside DSX OS: Modular Software for Scaling AI Factory Operations
At the heart of the NVIDIA DSX platform is DSX OS, a set of open, modular software components built for operating multi-tenant AI factories at scale. DSX OS standardizes how infrastructure teams provision clusters, coordinate workloads, and manage fleets across regions, turning fragmented tools into a cohesive operating layer. NVIDIA is releasing software that already runs its own DGX Cloud as open source, so ecosystem partners can reuse proven components instead of rebuilding them. According to NVIDIA, this approach eliminates months of custom development and moves operations from reactive alerting to automated remediation, with consistent runtime versions and fleet-wide visibility. By tying software, power management, and facilities data into a single architecture, DSX OS aims to improve tokens per watt, reduce token cost, and make AI factory software easier to integrate into existing infrastructure management stacks.
Maximizing Power Efficiency with DSX MaxLPS and Infrastructure Management
Power has become the primary constraint on AI infrastructure management, and DSX adds MaxLPS to improve efficiency inside the same power envelope. DSX MaxLPS combines 45-degrees-Celsius liquid cooling with in-rack performance-per-watt optimizations, so operators can run up to 40% more GPUs at their most energy-efficient operating point with minimal impact on workload performance. By treating power and grid behavior as a central part of the AI factory software stack instead of a separate facilities concern, DSX links energy use directly to AI output. This alignment extends across compute, networking, storage, cooling, and controls, so infrastructure builders can tune the entire environment for tokens per watt. The result is a blueprint for AI factories that delivers higher density and better efficiency without requiring new power feeds, supporting gigawatt-scale growth over time.
Simulation, Digital Twins, and Pre-Deployment Validation
A key promise of NVIDIA DSX is the ability to simulate an AI factory before any physical build-out. The platform is designed to integrate with digital twin tools, such as Vertiv’s simulations, so operators can validate power, cooling, and performance characteristics before installing a single rack. This model-driven approach turns AI factories into software-defined systems: architects adjust reference designs, test failure scenarios, and refine capacity plans in a virtual environment before committing to hardware and facilities changes. Jensen Huang summarizes the value by saying DSX lets teams “simulate the entire factory before you spend a dollar, [and] validate performance before a single rack is installed.” For enterprises, this reduces deployment risk, shortens time to revenue, and improves confidence that an AI factory will meet performance and reliability targets under real-world conditions.
Factory Operations Blueprint: Toward Autonomous Factory Operations
Beyond datacenter-scale AI factories, NVIDIA is targeting physical manufacturing with its Factory Operations Blueprint, codenamed FOX. This reference design defines how to build autonomous factory operations by unifying data from PLCs, SCADA, MES, and ERP systems into a single decision-making layer. FOX is not a product but an architecture that ingests live machine signals, quality data, and operational alerts, then feeds them into central AI models. NVIDIA Metropolis handles vision-based quality inspection, while NVIDIA Omniverse creates digital twins that mirror real production lines in simulation. Together, these tools support autonomous factory operations where AI optimizes workflows in real time, moving plants beyond isolated automation islands. For infrastructure builders, FOX shows how the NVIDIA DSX platform and AI factory software principles can extend from cloud-scale AI infrastructure management into on-premises, plant-wide intelligence systems.
