Physical AI Faces Costly Data Labeling Crisis

Robotics AI models train on fewer than 5,000 hours of open-source real-world interaction data despite over $10 billion in 2025 funding. Physical data requires costly collection, unlike web-scraped text. Scale AI leads real-world data collection backed by Meta's $14.3 billion stake. NVIDIA generates synthetic data, but quality remains a concern.