THE CRUNCH
Genesis AI, a Silicon Valley startup building a humanoid robot around a physical AI model, has raised a $500 million Series A led by Khosla Ventures, Premji Invest and Eclipse. CEO and co-founder Zhou Xian says the company has "cracked dexterous manipulation", and hosted a demo in which off-the-shelf robotic arms picked up small boxes, packed them into a larger one, folded, taped and labelled it, a workflow today's robots typically struggle with. The company's bet is vertical integration: it generates most of its own training data, builds its own generative AI model, and designs its own humanoid robot, Eno, all in-house.
The distinctive ingredient is force data. More than 100 workers worldwide wear sensor-equipped gloves and headset cameras, and their activities feed the training dataset. Genesis's robots wear the same gloves, and their hands are proportioned like human hands, which Xian credits for the system's sorting and wire manipulation abilities. Heavy investment in simulation, using a physics algorithm co-authored by adviser Minchen Li, generates millions of high-fidelity training scenarios.
The demo also showed the gap between lab and reliable automation. When Genesis re-ran the presentation, the robot fumbled a box and struggled to peel the tiny adhesive label off printer tape. A Genesis executive suggested such mistakes are useful "hiccup data" for improving the model.
Eno itself was not ready to demonstrate. Genesis says it has built 30 to 40 of the humanoids and has finalised the design of the mass-produced version, and would only say that a mass-manufacturing contractor had been chosen to start work next year in the US, with pilots under way at undisclosed customers. Xian names hardware as the real bottleneck: the dexterous hands currently last only a few weeks of regular work before needing replacement. He also plays down architecture debates, estimating that model design contributes "maybe 10%" of the final result, with data and training infrastructure doing the rest.


