Physical AI Stuck in GPT-2 Era - Unitree’s Valuation Paid Price

👁 116 reads

Unitree, China’s leading robot maker, lost nearly half its market value this week after debuting at a $66 billion valuation on China’s equivalent of the NASDAQ - a collapse analysts attribute not to hardware failures but to robots’ continuing inability to perform reliable, value-creating work.

The timing coincided with last week’s Actuate conference, a developer gathering focused on AI brains for robots, where the sector’s fundamental constraints were on full display. The event, organised by data tooling company Foxglove, has tripled in size since its 2023 launch and drew 1,500 attendees this year. A booth for infrastructure startup Avala promised to solve “the robotics data crisis” - shorthand for the scarcity of high-quality training data that continues to bottleneck model development.

Harry Mellsop, a founder of simulation-tooling startup Antioch, described physical AI as being in its “GPT-2 era,” a reference to the OpenAI model that predated ChatGPT. Closing that gap, he argues, will require more diverse datasets, better reinforcement learning setups, and GPUs optimised for ray tracing to generate high-fidelity simulations.

Autonomous vehicles remain the furthest along, partly because cars driven by humans generate relevant training data naturally, and partly because the core task - avoiding contact - is simpler than manipulating physical objects. Foxglove itself was founded by former employees of Cruise, General Motors’ self-driving unit. Both AV company Wayve and ride-share giant Uber have now launched humanoid robotics R&D labs, betting their machine-learning infrastructure translates across embodiments.

“I think you need to start in vehicles - manipulation robotics is like self-driving five years ago,” Wayve CEO Alex Kendall said. He argues it is too early to commit to any single hardware platform, favouring models that stay agnostic as sensors improve. Genesis AI CEO Théophile Gervet, whose company raised a $105 million seed round this year, took the opposing view, backing vertical integration.

Practitioner takeaway: For developers building with or on top of physical AI systems, the honest signal from Unitree’s valuation drop and the Actuate data-crisis conversation is that commercial deployment timelines remain uncertain. Investing in simulation pipelines and reinforcement learning tooling now - rather than betting on a single robot hardware vendor - is the approach Kendall and others are publicly advocating, and it mirrors the data-infrastructure lesson the LLM wave already taught.

// read moreWorld Models: How Genie 3 and Marble Teach AI Physical RealityState of AIAI Image Upscalers: Best Free and Paid ToolsAI ToolsGeneral Intuition in Talks to Raise at $6B ValuationMagazine

Kinetyra