Physical AI is currently experiencing explosive growth as venture capitalists pour billions into developing robot models inspired by large language models. This boom saw China's Unitree valued at an impressive $66 billion before its IPO, only to see its stock value halve within weeks due to its robots' limited ability to perform complex tasks.

The Actuate conference, which has tripled in size since 2023 with 1500 attendees, highlighted the urgency of addressing a significant hurdle: the lack of high-quality training data. Companies like Avala are offering solutions to this “data crisis,” with hopes that better simulation tools and diverse datasets will soon bridge the gap.

The automotive industry is leading the charge in physical AI, thanks to their ability to collect extensive real-world data from vehicles on the road. Autonomous vehicle companies such as Wayve and Uber have launched robotics labs focused on humanoid forms, seeing it as a potential future battleground for tech giants like Tesla with its Optimus robot.

However, not everyone is betting that general-purpose robots are the way forward. The CEO of Genesis AI, Théophile Gervet, argues against this approach and suggests focusing on specific verticals instead. This reflects the ongoing debate in the industry: should robots be designed for broad use or tailored to particular tasks?

The challenge lies in managing and utilizing vast amounts of data from sensors and simulations, with companies like Foxglove developing tools to facilitate this process more efficiently. Despite the progress, it may still be several years before we see a ChatGPT moment—where robots become as ubiquitous and useful as chatbots today.