Defining DSX: A Full-Stack Playbook for AI Factories
NVIDIA DSX is a full-stack AI infrastructure platform that treats the design, construction, and operation of AI factories as one unified software, hardware, and facilities challenge, combining reference designs, modular AI infrastructure software, digital twin simulation, and partner systems so enterprises can design, validate, deploy, and scale token-generating workloads with repeatable, standardized architectures. Instead of separate projects for chips, networks, data center facilities, and operations, DSX aligns compute, storage, networking, power, and cooling into a common architecture for AI factory infrastructure. NVIDIA positions it as a “complete playbook” for AI factories, spanning design, deployment, and ongoing operations. By giving infrastructure builders a consistent framework, DSX aims to reduce token cost, shorten time to first production, and improve operational reliability as AI becomes essential enterprise infrastructure. In effect, it turns AI factory planning into a software-led, model-based exercise rather than a series of point solutions.
How DSX Integrates Software, Reference Designs, and Partner Systems
At the core of the NVIDIA DSX platform is an integrated stack that joins modular AI infrastructure software, NVIDIA accelerated computing, and validated reference designs with partner technologies. DSX Reference Design provides generation-specific AI factory architectures that span compute, networking, storage, and cluster design, giving teams blueprints they can adapt instead of starting from scratch each build. These designs sit alongside facilities guidance for power distribution, cooling strategies, and controls, so infrastructure and IT teams work from the same plan. DSX also connects with partner ecosystems—such as facilities vendors and system integrators—to align their systems with NVIDIA’s co-designed architecture. This combination turns AI factory infrastructure into a more predictable, repeatable deployment pattern. According to NVIDIA, the goal is to convert available power into higher AI output and make AI factory infrastructure more standardized as token generation becomes a central enterprise workload.
DSX OS and MaxLPS: Turning Operations into AI Infrastructure Software
NVIDIA DSX OS brings open, modular AI infrastructure software to the operational layer, targeting lifecycle management, multi-tenant scheduling, platform services, and health automation for AI factories. Released as open source and derived from software used to operate NVIDIA DGX Cloud, DSX OS gives partners a ready-made operational stack instead of requiring months of custom development. DSX MaxLPS focuses on power and efficiency, combining 45-degrees-Celsius liquid cooling and in-rack power optimization to increase tokens per watt. NVIDIA states that “AI factories can run up to 40% more GPUs at peak energy efficiency within a fixed power budget, with minimal impact on inference workload performance.” Together, DSX OS and MaxLPS connect grid behavior, facility controls, and AI workloads into one coordinated system, shifting operations from reactive alerting toward automated remediation and consistent runtime management across regions for gigawatt-scale AI factory infrastructure.
Digital Twin Simulation: Omniverse DSX Blueprint and Vertiv SmartRun
A critical piece of the NVIDIA DSX platform is digital twin simulation, delivered through Omniverse DSX Blueprint and partners like Vertiv. Vertiv’s SmartRun overhead converged physical infrastructure is integrated as a configurable digital twin within Omniverse DSX Blueprint workflows, so teams can design, simulate, and validate power, cooling, and controls as one system before build-out. This model-based approach replaces traditional, document-heavy handoffs between engineering disciplines. By capturing configurations and dependencies in a virtual environment, infrastructure builders can test scenarios, reduce late-stage design changes, and lower integration risk. Vertiv describes this as the first phase of a broader AI factory digital twin roadmap intended to preserve engineering intent from early design through deployment and lifecycle optimization. In practice, digital twin simulation lets enterprises treat AI factory infrastructure as software—iterate in the virtual world, then deploy in the physical one with higher confidence and shorter time to operational readiness.
Why DSX Matters as AI Factories Become Standardized Infrastructure
As AI factories emerge as core infrastructure for token generation, enterprises need standardized, repeatable AI factory infrastructure rather than bespoke data center projects. DSX addresses this by aligning energy, chips, infrastructure, models, and applications into a single, co-designed architecture backed by open AI infrastructure software and reference designs. DSX OS gives operators fleet-wide visibility, consistent runtime management, and automated remediation for large-scale AI clusters, while MaxLPS connects efficiency tuning directly to power and cooling systems. Digital twin workflows in Omniverse DSX Blueprint further help teams validate designs before any racks are installed. By turning AI factory deployment into a unified software and simulation challenge, NVIDIA DSX lowers the barrier for infrastructure builders to scale AI factories quickly, operate them more efficiently, and respond to growing demand for token-based AI services without sacrificing reliability or resiliency.
