OpenAI bets against physical AI, two $500M raises, a payroll scandal, a robot makes a burger
OpenAI is betting against physical AI
For the second time in six years, OpenAI has shut down its dedicated robotics team.
The first time, in 2021, the reasoning was that physical AI was too hard to do well. This time the bet is almost exactly the opposite: OpenAI now thinks general-purpose frontier models do not need a dedicated robotics team because they already understand the world well enough to run robots directly. GPT-6 Astra, the model at the center of this, is being positioned as the robot brain. The same model that answers your questions is the one that tells a robot arm where to reach.
That is a large claim. Billions of dollars of investment are vote the opposite direction.
And, to be honest, the early returns are mixed at best.
Robots running pure frontier models without any dedicated physical-AI stack tend to be slower to react and worse at precise manipulation than robots trained on physical data. Boston Dynamics, Agility, Figure AI, Apptronik, 1X, and most of the Chinese players are betting the opposite way: that world models trained heavily on physical interaction will outperform general language models for anything that requires a body.
But scale is OpenAI's argument.
If Astra gets good enough fast enough, the dedicated physical-AI builders might find themselves running expensive training pipelines to close a gap that keeps shrinking. The same argument has been made about code, reasoning, and medicine. It has worked in at least some of those areas.
It might work here too. And it might not.
More info in the full story here ...
Two $500M raises, one week
Genesis AI, a California-based startup building the headless, wheeled Eno robot for factory floors, closed a $500M Series A at a $3.5B valuation. The company builds everything in-house: its GENE foundation model, dexterous hands with 20 degrees of freedom, matched data-capture gloves, and its own simulator.
The goal: human-level dexterity in robot hands. I've seen the hands in person, and they are very, very good.
Agile Robots did something structurally different.
The Munich company raised $500M at an $8.5B valuation, led by SoftBank and NVIDIA. But it did not build a robot and then look for a customer. It spent three years buying automation integrators, mobile robot companies, Franka (the robot arm maker), and thyssenkrupp's automation unit.
Now it's Agile ONE humanoid slots into a full-stack automation group with real revenue behind it. That is a different model from almost every other humanoid startup. And $8.5B is a pretty impressive valuation that has actual revenue behind it.
More info on Agile Robots' raise here
CEO borrowed $4.6M, then missed payroll
This one is not a good look.
Foundation Future Industries, the Eric Trump-backed startup claiming to build humanoid soldier robots with military contracts, missed payroll last month. One reason: its CEO had borrowed $4.6 million from company funds to buy a home. The military contract claims are falling apart under scrutiny. Investigations by NBC Bay Area and WIRED independently found that Foundation had not independently secured any military contracts.
The humanoid space attracts serious companies and serious capital. It also attracts this.
Also worth reading ...
Booster Robotics is betting that the next generation of humanoid robot developers is in school right now. With 3,000-plus robots shipped to 30-plus countries, Booster builds the hardware and leaves the intelligence to whoever buys it.
- I wrote about the model for Forbes: Humanoid Robots: This Company Sells The Body, You Supply A Brain.
- And HumanoidMegaHub has more on the education angle: Booster Robotics wants to turn kids into humanoid robot developers.
Also: European humanoid funding is not slowing down. Munich-based Vesoma is seeking €50 million, according to Sifted. No product details yet, but another European entrant positioning ahead of the Humanoid Technology Expo wave.
Quick hits ...
- Samsung's Future Robotics Task Force is testing a 0.5mm electronic skin film on robot hands, targeting commercial production in 2027. The practical value: robots that can feel what they are holding, adjust grip pressure, and detect slip. More: HumanoidMegaHub
- Neura Robotics and E.ON have partnered to deploy the 4NE-1 humanoid across European electrical infrastructure, with field trials at eastern Germany's largest electricity distribution network starting in 2027. More: HumanoidMegaHub
- AGIBOT's A3 Ultra can climb stairs, squeeze through narrow gaps, and adapt movement in real time. This is running in live commercial deployments, not demos. More: HumanoidMegaHub
- Tau Robotics put its humanoid to work making a burger. I want one too! Check out the video here
Number of the day: 300,000
JPMorgan analysts say the US will ship 300,000 to 400,000 humanoid robots by 2030, rising to 1.5 million by 2035. The bottleneck they identified:
- Not demand
- Not technology
- Production capacity
That is a manufacturing problem. It is the same constraint that slowed EV adoption for years. And it is the reason both Genesis AI and Agile Robots are spending their raises on scale rather than more research.
The demand is coming. The question is who can build fast enough to meet it.
Written by
I tell stories with data and build startups. I've raised angel and VC funds, led software companies and startups, and chronicled the rise of the mobile economy. Currently I write for Forbes, host several podcasts, and am starting up HumanoidMegaHub. I was TUNE's mobile economist and Singular's VP of Insights, built the research team at VentureBeat, and managed teams creating software for partners like Intel and Disney. In past lives I built websites, e-learning platforms, and mobile apps, and, a long time ago, was named to Folio's top 100 of the media industry's "most innovative entrepreneurs and market shaker-uppers.”