NVIDIA GTC: The Virtual World Opening the Physical AI Era

At NVIDIA GTC, a turning point was presented in which physical AI moves beyond single demos to expand into enterprise workloads across robots, vehicles, and factories. At the core are new frontier models such as NVIDIA Cosmos 3, Isaac GR00T N1.7, and Alpamayo 1.5, backed by simulation and data generation infrastructure.

OpenUSD was highlighted as a common language that ties together CAD data, simulation assets, and real-world telemetry into a single physically accurate scene. This allows the pipeline from design through validation to deployment to be built first inside a virtual environment.

The key announcements condense into two blueprints.

• NVIDIA Physical AI Data Factory Blueprint: an open reference architecture that generates large-scale, high-quality training data from limited real-world data
• NVIDIA Omniverse DSX Blueprint: a blueprint that unifies an AI factory into a single digital twin, simulating thermals, power, networking, and mechanical systems

On the data side, it's noted that real-world data is no longer a moat. Instead, the bottleneck is not the data itself but the entire data factory where collection, simulation, and evaluation are siloed, and this blueprint ties data curation, augmentation, and evaluation into a single flow built on NVIDIA Cosmos and OSMO.

This approach has also been extended to cloud platforms such as Microsoft Azure and Nebius. With these two platforms offering the blueprint first, a direction is set to turn world-scale compute into a turnkey data production engine.

In the design and deployment segment, the emphasis was on converting CAD to OpenUSD and optimizing simulation-ready assets with the NVIDIA Omniverse Kit SDK and NVIDIA Isaac Sim. FANUC and Fauna Robotics are using this CAD-to-OpenUSD workflow to speed up the design and validation of robotic systems.

In manufacturing and logistics, the view that the factory itself is a robotic system took center stage. The NVIDIA Mega Omniverse Blueprint was presented as a reference architecture for designing, testing, and optimizing robot fleets and AI agents inside a facility digital twin before actual equipment is installed, and KION is using it to build a large-scale warehouse digital twin for training and validating an autonomous forklift fleet for GXO.

On the hardware and ecosystem front as well, the real-world application of physical AI is expanding. ABB Robotics, FANUC, KUKA, and Yaskawa, with a combined installed base of over 2 million robots, are using NVIDIA Omniverse and the NVIDIA Isaac simulation framework to validate complex robotic applications and production lines, and are embedding NVIDIA Jetson modules in controllers to perform real-time AI inference.

In developing robot brains, FieldAI and Skild AI are combining NVIDIA Cosmos world models with Isaac simulation to carry out data generation and policy validation. Ultimately, the message from this GTC is clear: compute is data, and the scaling of physical AI depends not on collecting more of the real world, but on the ability to generate it faster and more precisely in the virtual world.