Summary
In a Bloomberg Technology interview, Nvidia VP Deepu outlines the company’s strategy for making humanoid robots safe enough to operate alongside humans — and explains why the field has not yet had its “ChatGPT moment.” Nvidia does not build robots itself; instead it supplies the core compute and software platform used by more than 10,000 companies and 2.5 million developers through its Nvidia Jetson edge compute line.
Deepu argues that humanoid robots face a fundamentally different accuracy bar than digital AI systems. When a user asks Claude to summarize a document, a human reviews the output. When a robot carries a heavy object past a person, there is no review loop — which means accuracy requirements range from nine to fifteen nines depending on the application. Current robots handle this by defaulting to safe distances and conservative motion, but the goal is genuine proximity operation. Safety, in Nvidia’s view, must be designed from the chip level upward through the OS and application layers, and must ultimately be explainable and certifiable by third parties — not just self-reported.
The conversation covers Nvidia’s “three computer” robotics framework: one computer for training the robot brain, one for simulation and testing via the Omniverse platform, and one embedded in the robot for real-time edge inference. Deepu references Nvidia’s recently announced HALO system as a key step toward giving humanoid robots layered situational awareness, and identifies general-purpose brain accuracy — not hardware or mechanics — as the primary remaining barrier to widespread humanoid deployment.
📺 Source: Bloomberg Technology · Published June 25, 2026
🏷️ Format: Interview
