Defining the AI Factory Operating System
An AI factory operating system is a full-stack control layer that coordinates chips, data center infrastructure, software, and AI workloads so enterprises can treat intelligence generation as a factory-scale process rather than isolated model deployments. NVIDIA’s new DSX platform embodies this idea by aligning computing, networking, storage, facilities, and partner technologies into a single architecture. Instead of focusing on one model or cluster at a time, DSX centers operations on token production, connecting power, cooling, and resource scheduling directly to AI output. NVIDIA describes AI as “essential infrastructure” and DSX as the playbook for designing, simulating, building, and operating AI factories. By offering shared reference designs and open components, it aims to shorten the path from planning to first production while improving reliability and the ratio of tokens per watt across multi-tenant environments.
Inside the NVIDIA DSX Platform: From Chips to Facilities
The NVIDIA DSX platform combines modular software libraries, APIs, reference designs, and accelerated computing platforms into one framework for AI infrastructure scaling. It spans the entire five-layer stack of energy, chips, infrastructure, models, and applications, binding them through a co-designed architecture. In practice, this means data center teams can plan compute, networking, storage, facility design, power, cooling, controls, simulation, and operations as one system instead of separate projects. Jensen Huang describes the intent clearly: “With the DSX platform, 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.” That focus on simulation and repeatable design turns each deployment into a standard “AI factory” unit, ready to be rolled out rapidly across sites and tenants.
DSX OS and MaxLPS: Open, Modular Control for AI Output
At the core of DSX is the NVIDIA DSX OS platform, an open source, modular OS for autonomous factory operations at cloud scale. DSX OS packages components NVIDIA uses in its own DGX Cloud, now released so partners can build AI services without spending months rewriting infrastructure software. It is designed to plug into existing platforms, connecting chips, servers, building management systems, and AI services into a coordinated control plane. Alongside DSX OS, the DSX MaxLPS suite focuses on power efficiency, pairing 45-degrees-Celsius liquid cooling with in-rack optimizations. According to NVIDIA, this lets operators run up to 40% more GPUs at their most energy-efficient point within a fixed power budget, with minimal impact on inference performance. Together, DSX OS and MaxLPS push AI factory economics toward more tokens per watt and lower token cost, while improving resilience for continuous workloads.
Factory Operations Blueprint: Toward Autonomous Factory Systems
NVIDIA’s Factory Operations Blueprint, codenamed FOX, extends the DSX philosophy into industrial manufacturing, where fragmented control stacks limit plant-wide intelligence. Traditional factories rely on isolated PLC, SCADA, MES, and ERP systems that rarely share a unified context, making root cause analysis slow and predictive maintenance difficult. FOX defines an architectural guide rather than a single product, creating a unified decision-making layer that ingests live machine signals, quality systems, and operational alerts into a central AI model. Built around NVIDIA’s existing hardware and software stacks, it connects data ingestion from legacy PLCs and modern IoT sensors to frameworks like NVIDIA Metropolis for automated vision-based quality inspection. The result is a feedback loop between digital simulation and physical operations, aiming to move factories from task-specific automation toward autonomous factory operations where AI optimizes workflows and maintenance in near real time.
Digital Twin Simulation with Vertiv SmartRun and DSX
Digital twin simulation is a critical piece of DSX’s promise to validate AI factories before build-out, and Vertiv’s SmartRun integration shows how this works in practice. Vertiv SmartRun, an overhead converged physical infrastructure system, is now modeled as a configurable digital twin inside NVIDIA Omniverse DSX Blueprint workflows. Infrastructure teams can design, simulate, and validate power, cooling, and controls as a single system before any physical deployment. This model-based approach replaces document-driven, siloed handoffs that slow projects and introduce integration risk. By capturing configurations and dependencies virtually, enterprises can reduce late-stage design changes, preserve engineering intent from early planning through commissioning, and better coordinate cross-functional teams. As Vertiv advances its multi-phase AI factory digital twin roadmap, DSX gains a tangible path for turning each new generation of accelerated compute into repeatable, simulation-ready AI factory infrastructure.
