Nvidia site visit

Physical AI, which is where AI intersects with the physical world, entails completely new trillion dollar TAM opportunities with use cases such as autonomous vehicles (AVs) and all forms of autonomous & intelligent robots.

‍

‍

The physical world will be totally simulated and we have no doubt Nvidia will be leading this new wave as well, with its solutions like Omniverse and Cosmos which form the operating system on top of its existing full stack AI infrastructure platform.

In any autonomous robotics application, there will be a need for 3 separate Nvidia supercomputers:

1. in training of physically grounded, 3D world foundation models (centralized, datacenters)

2. for ongoing robotics learning through simulation via RL (Cosmos)

3. a localized brain for each deployed robot (Jetson/Project Digits) for perception/observation and actions (localized, edge AI)

The compute requirement (Nvidia’s GB200 chip below) for these new robotics applications will be at least 10x - 100x those of current traditional autoregressive LLM applications, once we overcome the initial high quality data problem we currently have with robots via human captured data pipelines.