From Experimental Labs to AI Factories at Scale
NVIDIA DSX is an AI factory operating system and full-stack platform that standardizes how enterprises design, simulate, deploy, and operate large-scale infrastructure for generating AI tokens as a reliable production service. Rather than treating AI as a collection of experimental clusters, the NVIDIA DSX platform aligns chips, systems, software, facilities, and partner technologies around a common architecture, so organizations can plan AI capacity with the same discipline they apply to traditional data centers. Announced at NVIDIA GTC Taipei, DSX combines open source, modular software libraries, APIs, reference designs, accelerated computing platforms, and ecosystem technologies to reduce token cost and shorten time to first production. Jensen Huang described the intent clearly: “We’re not just shipping chips — we’re giving every infrastructure builder a complete playbook to build AI factories,” underlining the shift from one-off builds to standardized, repeatable AI factory designs.
Inside the NVIDIA DSX Platform: A Full AI Factory Stack
The NVIDIA DSX platform is designed as a full-stack foundation for enterprise AI infrastructure, covering compute, networking, storage, power, cooling, controls, simulation, and ongoing operations. It treats AI factories as multi-layer systems that must coordinate energy, chips, infrastructure, models, and applications to maximize tokens produced per watt. By tying together NVIDIA accelerated computing platforms with partner technologies and reference architectures, DSX gives infrastructure teams a coherent path from whiteboard to live service. A core promise is the ability to simulate the entire factory before deployment, validate performance before any rack is installed, and then run it as a standardized asset instead of a custom science project. This moves token generation scale planning closer to capacity management in conventional data centers, with predictable layouts, known performance envelopes, and documented practices that can be replicated across sites.
DSX OS: The AI Factory Operating System
At the heart of this shift is DSX OS, open source, modular software built for operating and scaling multi-tenant AI factories. DSX OS targets the complex network of components involved in large AI workloads: GPU clusters, storage, networking, building management systems, cooling, power distribution, grid interactions, and the AI platforms running on top. According to NVIDIA, these components are aligned around a common architecture to achieve three outcomes: faster time to revenue, better efficiency, and higher reliability. By releasing software that also runs NVIDIA DGX Cloud, DSX OS lets partners build AI services without months of custom development. Power-aware operation is built in: DSX software can support running up to 40% more GPUs at peak energy efficiency within a fixed power budget, with minimal impact on inference performance, directly improving tokens per watt and overall token generation economics.
Maximizing Tokens per Watt with DSX MaxLPS
Scaling enterprise AI infrastructure now hinges on power efficiency as much as raw performance. DSX MaxLPS, a key software and systems component of the NVIDIA DSX platform, focuses on maximizing token performance per megawatt within fixed energy limits. It combines 45-degrees-Celsius liquid cooling with in-rack technologies tuned to performance per watt, so operators can convert available power into higher AI output instead of leaving stranded capacity. NVIDIA states that this design lets AI factories run up to 40% more GPUs at their most energy-efficient operating point with minimal impact on workload performance. Integrated with DSX OS, MaxLPS elevates power from a facilities constraint to a programmable resource, allowing teams to plan token generation scale, manage thermal envelopes, and respond to grid conditions as part of normal operations rather than emergency optimization exercises.
Factory Operations Blueprint: Template for Autonomous AI Factories
Beyond data centers, NVIDIA is also targeting physical manufacturing with its Factory Operations Blueprint, codenamed FOX, which mirrors DSX’s goal of turning fragmented systems into unified, AI-driven factories. FOX is not a single product but a reference design that integrates Programmable Logic Controllers, SCADA, MES, and ERP into a unified decision layer. It uses NVIDIA Metropolis for vision-based quality inspection and NVIDIA Omniverse to build digital twins that connect live factory signals with simulated workflows. This architecture creates a feedback loop where AI can manage and optimize complex processes in real time, bringing plant-wide intelligence to historically siloed environments. For enterprises, FOX plays a similar role to DSX OS: a template of standardized components and patterns that makes it possible to deploy autonomous, AI-managed factory systems with the predictability and repeatability of modern data centers.
