What Is NVIDIA DSX and Why It Matters
NVIDIA DSX is a full-stack platform that combines software, reference designs, simulation tools, and partner systems into a unified operating framework for AI factory infrastructure, helping organizations scale token generation efficiently from initial design to live operations. In practical terms, NVIDIA DSX operating system concepts align chips, systems, facilities, and AI workloads around a co-designed architecture, rather than leaving infrastructure builders to assemble point tools. NVIDIA describes DSX as a “complete playbook” for AI factories, spanning compute, networking, storage, power, cooling, controls, and operations. DSX aims to improve tokens per watt, shorten time to first production, and increase reliability at scale. As AI becomes essential infrastructure, DSX turns data centers into AI factories, where tokens—outputs of large models—are produced with clear attention to energy limits, operational complexity, and multi-tenant environments.
Inside DSX OS: Modular Software for AI Factory Operations
DSX OS is the open, modular software layer of the NVIDIA DSX platform, designed to operate multi-tenant AI factories at scale. It packages open source components and NVIDIA technologies used in NVIDIA DGX Cloud and releases them for infrastructure builders who need a consistent operations stack instead of bespoke integrations. According to NVIDIA, DSX OS is built for the five-layer stack of energy, chips, infrastructure, models, and applications, making power behavior part of the platform rather than a separate facilities concern. The software coordinates chips, systems, cooling, building controls, and AI services to improve tokens per watt and lower token cost. DSX MaxLPS complements this by combining 45-degrees-Celsius liquid cooling with in-rack optimizations so operators can run up to 40% more GPUs at their most energy-efficient point with minimal impact on workload performance.
Digital Twin Simulation with Vertiv SmartRun and Omniverse
A central promise of AI factory infrastructure is the ability to simulate everything before building anything, and NVIDIA DSX integrates digital twin simulation to support that goal. Vertiv’s SmartRun digital twin is integrated into the NVIDIA Omniverse DSX Blueprint, giving infrastructure teams a way to design, simulate, and validate power, cooling, and physical layouts as a single system. This digital twin simulation replaces traditional document-based planning with a model-based approach, preserving engineering intent from early configuration through deployment and lifecycle optimization. Infrastructure builders can test configurations, capacity limits, and failure scenarios virtually, then roll those validated designs into DSX-powered AI factories. As Vertiv expands this roadmap, the SmartRun digital twin becomes a repeatable foundation for AI factory infrastructure, helping close the gap between rapid GPU innovation and physical readiness while reducing late-stage design changes and integration risk.
Factory Operations Blueprint: A Reference Design for Autonomous Plants
Beyond data centers, NVIDIA DSX connects to manufacturing through the Factory Operations Blueprint, codenamed FOX, which acts as a reference design for autonomous factory systems. Modern plants run separate stacks—PLCs for machine control, SCADA for process monitoring, MES for workflows, and ERP for business logistics—often without a unified view. The factory operations blueprint proposes a single decision-making layer that unifies these systems and feeds a central AI model. It defines data ingestion paths from legacy PLCs and modern IoT sensors, uses NVIDIA Metropolis for vision-based quality inspection, and closes the loop between digital simulation and physical operations. With this factory operations blueprint, infrastructure builders can implement plant-wide intelligence instead of task-specific automation, enabling predictive and prescriptive maintenance, faster root cause analysis, and real-time optimization of production lines that align with AI factory infrastructure principles.
A Full-Stack Platform for Infrastructure Builders, Not Point Tools
NVIDIA DSX is shaped for infrastructure builders who need an end-to-end AI factory infrastructure, not a set of disconnected products. The platform gathers open source DSX OS components, NVIDIA accelerated computing, reference designs like the factory operations blueprint, and partner systems such as Vertiv SmartRun into a single, co-designed architecture. This means teams can design AI factories, simulate them through digital twin tools, validate performance and power behavior, and then operate them with consistent software and controls. NVIDIA states that with DSX “you can simulate the entire factory before you spend a dollar, validate performance before a single rack is installed and operate with the kind of reliability that production AI demands.” For organizations building AI factories, DSX offers a modular but integrated path from concept to continuous operations, with token generation economics at its core.
