Defining DSX: From Data Centers to AI Factories
NVIDIA DSX is an AI factory operating system: an open, modular platform that standardizes how chips, software, and facilities work together so data centers can design, simulate, and operate large-scale token-generation infrastructure more efficiently. Instead of treating AI deployments as one-off projects, DSX provides a repeatable blueprint that aligns energy, compute, networking, storage, and building systems into a coherent AI factory. The platform covers the full lifecycle, from early design to ongoing operations, with an emphasis on improving tokens per watt and lowering the cost of intelligence across large fleets of GPUs. By turning infrastructure into a common architecture rather than a collection of custom integrations, NVIDIA aims to shorten time to first production and increase operational reliability for enterprises that need to scale AI infrastructure quickly.
Inside the NVIDIA DSX Platform: A Full-Stack AI Factory Playbook
The NVIDIA DSX platform combines software libraries, APIs, reference designs, accelerated computing platforms, and partner technologies into a single, modular stack. It is designed around the idea that AI infrastructure must be co-architected across five layers: energy, chips, infrastructure, models, and applications. DSX brings these layers together, aligning compute, networking, storage, facility design, power, cooling, controls, simulation, and operations. This full-stack approach is not limited to NVIDIA hardware; DSX OS components are open source and extensible, so operators can integrate them into existing platforms while standardizing toward a common architecture for AI factory operations. According to NVIDIA, this alignment enables AI factories to run up to 40% more GPUs at their most energy-efficient operating point within a fixed power budget, with minimal impact on inference workload performance, which goes straight to improving tokens per watt.
DSX OS and MaxLPS: Standardizing AI Factory Operations at Scale
At the heart of DSX is DSX OS, an open, modular software stack created specifically for operating multi-tenant AI factories at scale. DSX OS packages the software NVIDIA uses to run its own DGX Cloud infrastructure and releases it to the broader ecosystem, so partners do not need to rebuild core operational services from scratch. The stack coordinates resource scheduling, observability, and integration with facility systems such as power and cooling, aiming to improve reliability and resiliency for continuous, large-scale workloads. Complementing it, DSX MaxLPS brings 45-degrees-Celsius liquid cooling and in-rack technologies focused on maximizing token performance per megawatt. Together, DSX OS and MaxLPS help operators convert fixed power budgets into higher AI output, directly supporting AI infrastructure scaling while simplifying how new capacity is brought online and managed in production.
Factory Operations Blueprint: A Reference for Autonomous AI Factories
While DSX targets AI data centers, NVIDIA’s Factory Operations Blueprint (codenamed FOX) shows how similar principles can drive autonomous factory operations on the industrial floor. Most plants run a patchwork of PLC, SCADA, MES, and ERP systems that seldom integrate well, which blocks plant-wide intelligence and makes root-cause analysis slow and manual. FOX acts as an architectural guide rather than a product, defining a unified decision-making layer that ingests live machine signals, quality control outputs, and operational alerts into centralized AI models. Built around NVIDIA hardware and software, including tools like Metropolis for vision AI, the blueprint creates a feedback loop between digital simulation and physical operations. This structure pushes factories beyond task-level automation toward AI-managed workflows, where models continuously optimize throughput, quality, and maintenance, echoing the same control-loop mindset DSX brings to AI infrastructure operations.
Omniverse DSX and Vertiv SmartRun: Simulating AI Factories Before Build-Out
Simulation is where DSX’s AI factory story becomes tangible. Through Omniverse DSX, infrastructure teams can create digital twins of data center systems and test configurations before any physical build-out. Vertiv’s SmartRun digital twin, integrated with the NVIDIA Omniverse DSX Blueprint, illustrates how partner systems plug into this model-based workflow. SmartRun represents Vertiv’s overhead converged physical infrastructure as a configurable digital twin, allowing power, cooling, and controls to be designed, simulated, and validated as one system. This reduces late-stage design changes and integration risk while improving coordination across teams, which is critical as AI deployments move to higher densities and larger capacities. By shifting planning from documents to live digital twins, DSX and its ecosystem of infrastructure simulation tools make AI factories more repeatable, predictable, and ready for rapid scaling of token-generation workloads.
