NVIDIA Defines Physical AI at SIGGRAPH: EB-Scale Data, OpenUSD and Sim-to-Real
During SIGGRAPH 2026's NVIDIA Physical AI Day, NVIDIA systematically showcased its latest roadmap for physical AI and embodied intelligence. The core message was clear: the development of physical AI relies heavily on simulation, and its underlying infrastructure and data acquisition paradigms are being completely redefined.
Confirmed
Standards and Foundation: OpenUSD and Synthetic Data
NVIDIA is aggressively promoting OpenUSD as the universal standard for 3D content, simulation, and physical AI workflows, urging the industry to adopt it and join the Alliance for OpenUSD. Officials emphasized that physical AI starts with simulation, with OpenUSD and synthetic data forming a solid foundation. NVIDIA is also providing free OpenUSD resources to lower development barriers.
Shift in Data Generation Paradigm
Training for embodied and physical AI is shifting from slow physical data collection to scalable, generalizable data generation. NVIDIA noted that humanoid robots and world models might require EB-scale training data—about 1000 times more than autonomous driving. To overcome this bottleneck, the company proposed using World Foundation Models to drive data generation for physical AI.
Sim-to-Real Applications
Lightwheel AI demonstrated a practical embodied robotics case: a robot was first trained in the Isaac Lab Arena simulation and subsequently deployed in a real operating room to perform tasks, signaling the accelerated real-world deployment of simulation tech in complex scenarios like healthcare.
Why it matters
NVIDIA's initiatives indicate that the biggest bottleneck for embodied intelligence is acquiring massive amounts of high-quality training data. By establishing a unified 3D data standard (OpenUSD) and leveraging compute and foundation models to mass-generate simulation data, NVIDIA is attempting to build a simulation-centric foundational infrastructure ecosystem for the next generation of physical AI.
2026-07-22 ~ 2026-07-23 · 6 related posts
Primary sources
- [source] NVIDIA pitches World Foundation Models as a way to scale physical AI data generation — MonaJalal_ · 2026-07-22
- NVIDIA says physical AI starts in simulation with OpenUSD and synthetic data — MonaJalal_ · 2026-07-22
- [source] NVIDIA pushes OpenUSD as the common layer for simulation and physical AI — MonaJalal_ · 2026-07-22
- NVIDIA Physical AI Day: Talk on SimReady 3D Worlds — MonaJalal_ · 2026-07-22
- [source] NVIDIA demo shows Isaac Lab robots moving from simulation into surgery tasks — MonaJalal_ · 2026-07-22
- NVIDIA says physical AI needs exabyte-scale simulation data at SIGGRAPH 2026 — jonstephens85 · 2026-07-23