NVIDIA details agent-assisted workflow for AI factory digital twins

NVIDIA said developers can use an agent-assisted workflow to assemble and revise AI factory digital twins from site plans, equipment assets and engineering requirements, using its Omniverse DSX Blueprint and OpenUSD-based tooling. The Oct. 21, 2026 post describes a process intended to help teams inspect design alternatives, prepare scenes for engineering review and connect the digital twin to downstream simulation.

Evidence and context

In the blog post published Oct. 21, 2026, NVIDIA said the Omniverse DSX Blueprint, part of NVIDIA DSX Sim, provides reference digital twin assets, sample scripts and workflows for developing AI factory digital twins. The company said the blueprint is built on OpenUSD and NVIDIA Omniverse libraries and is designed to help developers assemble facility and equipment models with SimReady assets while linking them to engineering simulation tools.

The post describes a workflow in which a frontier AI model operates inside an agent harness that manages context, instructions and tool calls. NVIDIA said developers can guide agents to bring together design inputs, assemble and revise OpenUSD scenes, and inspect the results. The company said the approach is intended to support editable campus and hall models that can be reviewed before engineering sign-off and simulation.

NVIDIA illustrated the workflow with an AI factory campus example that combines DSX 3D models, architectural references, site plans, requirements documents and geographic data. The inputs included data hall, utility and cooling assets; images of the NVIDIA Santa Clara Headquarters building, Voyager; PDFs describing campus arrangement; indexed drawings and basis-of-design records; and USGS aerial imagery and elevation data for the selected site, according to the post. NVIDIA said those sources help teams trace design choices back to their origin, while noting that dimensional accuracy still depends on checking drawings, specifications and site constraints.

How the workflow is intended to work

NVIDIA said the campus example was built by combining existing DSX reference-design geometry with new architectural elements in additive USD layers, keeping changes separate from source assets. The company said its team used USD Content Agents to help generate a solar canopy’s geometry and materials, then refined the result using Voyager references. The canopy was described as conceptual and still requiring engineering review for dimensions, structure and simulation suitability.

The post also says developers can use Systems Modeling Language to connect requirements to system elements, and that an agent may query those relationships where tooling exposes them through an API, MCP or another supported interface. NVIDIA said drawings and CAD would provide dimensions and placement references, while equipment catalogs would supply parameters for analysis. The handoff, in this framing, is a proposed design revision accompanied by the affected requirements, equipment and follow-up studies.

For simulation preparation, NVIDIA said OpenUSD brings building geometry, equipment assets and associated data into a shared scene, while USD attributes and custom schemas can preserve links to engineering records and requirements. The company also pointed to SimReady Foundation and the SimReady Central Library on Hugging Face, which it described as a free library of more than 2,000 simulation-ready OpenUSD assets. NVIDIA said these resources can reduce asset-preparation work by providing OEM racks and power and cooling equipment.

Capacity-planning example and limits

One proposed extension in the post is a fixed-power capacity-planning study for an AI factory hall. NVIDIA said an illustrative Vera Rubin NVL72 inference comparison contrasts a baseline using MaxP, described as the highest static GPU-level power setting, with DSX MaxLPS, an AI factory design and operating framework that coordinates facility design, dynamic power allocation and performance-per-watt optimization.

Normalized to a 1 MW power basis and assuming a power usage effectiveness of 1.1, the post says the illustrative scenario shows a DSX MaxLPS capacity of 400 GPUs and aggregate token throughput of 1.35 times the MaxP baseline. NVIDIA said those figures are reference assumptions for the proposed study, not results generated by the campus model. The company also said Vera Rubin testing remains in progress.

The post says an engineering workflow would still require design reviews, solver connections, equipment properties, system connections and boundary conditions before simulation. NVIDIA said the objective is to evaluate electrical loading and cooling behavior, then feed the findings back into the editable design for another iteration. The company stressed that each finding should identify the tested design revision and distinguish calculated results from stored results or AI approximations.

Related tools and partner applications

NVIDIA also described hardware choices for different stages of the workflow, including DGX Spark for personal agent development, RTX PRO Blackwell Workstation Edition GPUs for interactive design and simulation, DGX Station for larger local AI workloads, and RTX PRO Server for shared development and simulation. It cited its RTX PRO AI Factory reference architecture, which specifies eight RTX PRO 6000 Blackwell Server Edition GPUs per server and groups four servers into a scalable unit.

The post points to application providers working across requirements management, product data and simulation. NVIDIA highlighted Dassault Systèmes’ CATIA Magic and A2A support, PTC’s Windchill and Omniverse libraries integration, Cadence Reality Digital Twin Platform for power and cooling studies, and Siemens’ Simcenter STAR-CCM+ for fluid-flow and heat-transfer analysis. NVIDIA said these tools are intended to keep requirements, equipment data and physical-behavior studies connected throughout design iterations.

Recent related NVIDIA posts add context to the company’s broader DSX push. On Oct. 21, 2026, NVIDIA also said it introduced DSX POD AI Factories to accelerate Europe’s infrastructure deployment, following an Oct. 7, 2026 post on validating AI factory changes with digital twins and AI agents and a March 16, 2026 announcement of the Vera Rubin DSX AI Factory Reference Design and Omniverse DSX Digital Twin Blueprint.