Goldman Sachs: AI robots may need 10 million hours of interaction data for production

General-purpose AI robots could require about 10 million hours of high-quality multimodal physical interaction data to reach production-level capabilities, according to a Goldman Sachs report. Progress in physical AI may be two to four times faster than autonomous driving, but 3D spatial perception, inference latency and inference economics remain key bottlenecks.

• Goldman Sachs estimates roughly 10 million hours of high-quality physical interaction data
• Physical AI progress may be 2–4 times faster than autonomous driving development
• Three bottlenecks: 3D spatial perception, inference latency and inference economics
• Industrial manufacturing demands six-sigma reliability and microsecond repeatability