Physical AI is artificial intelligence that perceives, reasons about, and acts in the physical world — controlling robots, vehicles, and industrial machines rather than generating text or images. Where a language model predicts the next word, a physical AI model predicts the next motor action: how to grip a box, balance on two legs, or steer a truck through traffic.
The term covers several overlapping approaches. Embodied AI emphasizes learning through a body's interaction with its environment. Robot foundation models — the approach behind Physical Intelligence's π-series and Skild AI's Skild Brain — train one large model across many robot types and tasks, betting that generality wins like it did in language. Humanoid robotics puts that intelligence in a human-shaped body so it can work in spaces built for people.
Why Did Physical AI Funding Explode in 2025-2026?
Three forces converged. First, foundation-model techniques from language AI transferred to robotics: models trained on massive demonstration datasets began generalizing across tasks they were never explicitly taught. Second, hardware costs collapsed — Unitree's G1 humanoid lists at $16,000, a price that made fleets affordable for training-data collection. Third, labor economics: warehouse, manufacturing, and logistics operators facing chronic staffing shortages became willing pilot customers.
The funding followed. Skild AI reached a $14 billion valuation with a $1.4B SoftBank-led Series C in January 2026 — eighteen months after its $1.5B Series A. Physical Intelligence hit $5.6 billion with a CapitalG-led $600M round. Figure AI raised over $1B at a $39 billion post-money valuation. In China, Galbot's $300M+ round at $3B set a national embodied-AI record in December 2025, and Unitree filed for a STAR-market IPO at a reported ~$7B target valuation.
Who Are the Main Physical AI Investors?
A distinct investor set has formed around the sector. Khosla Ventures led rounds at Waabi ($750M Series C) and Field AI ($405M). NVIDIA's NVentures appears across Skild, Nuro, Waabi, Dyna Robotics, and Field AI — the chipmaker funding demand for its own robotics compute. Amazon's Industrial Innovation Fund backs Agility Robotics, Skild, and Dyna. Bezos Expeditions anchors Physical Intelligence and Field AI. In China, state-linked capital dominates: China Mobile's funds led both Unitree's Series C and Galbot's Series B, while automakers NIO, Geely, and BAIC fund humanoid makers as future factory customers.
How Is Physical AI Different From Traditional Robotics?
Traditional industrial robots execute pre-programmed motions in caged, structured environments — the same weld, thousands of times. Physical AI systems learn from demonstration and generalize: the same model that folds laundry can learn to load a dishwasher. That flexibility is why investors price these companies like AI labs (10-50x forward revenue) rather than hardware makers (2-5x), and why the sector's defining question is whether general-purpose robots can reach reliability levels that justify those multiples before the capital cycle turns.