· via TechCrunch
Genesis AI raises $500M Series A for dexterous humanoid robots trained with sensor gloves
Genesis AI has raised a $500 million Series A led by Khosla Ventures to pursue dexterous robotic manipulation, training its models on force data from sensor gloves worn by human workers.

Genesis AI lands $500 million for physical AI
Genesis AI, a Silicon Valley startup building humanoid robots and the AI models that run them, has raised a $500 million Series A led by Khosla Ventures, Premji Invest and Eclipse, according to TechCrunch. The funding arrives with an unusually direct claim from co-founder and CEO Zhou Xian, who told the publication his company has "cracked dexterous manipulation" — the ability of machines to handle objects the way people do.
Founded in 2024 by Xian and company president Théophile Gervet, who met as graduate students at Carnegie Mellon, Genesis previously announced a $105 million seed round in July 2025. Its strategy is vertical integration: the company generates most of its own training data, develops its own generative AI model, and designs its humanoid hardware in tandem, an approach Xian frames as shipping the best possible robot as an end-to-end product.
A demo that impressed, then stumbled
At Genesis's facility, TechCrunch watched the company's physical AI model drive a pair of off-the-shelf robotic arms through a packing workflow: lifting two small cardboard boxes, scanning them, placing them inside a larger box, folding it shut, taping it, and peeling a label off a strip of printer tape to apply to the finished package. Tasks of that complexity remain genuinely difficult for today's robots.
When the demonstration was repeated, the robot dropped a box and struggled to peel the small adhesive label loose. Genesis treats such failures as raw material — an executive called them "hiccup data" — the kind of edge cases that can be fed back into the model to improve it. As TechCrunch observes, the distance between an impressive lab demo and automation a paying customer can depend on is precisely the gap that Genesis, and the wider humanoid ecosystem, must close to avoid another robot winter.
Gloves as the data engine
The company's main differentiator, in TechCrunch's telling, is its focus on dexterity and the data needed to achieve it, which sets it apart from competitors such as Figure and Agility. More than 100 workers around the world wear sensor-equipped gloves and headset cameras while doing their jobs, and the force data those gloves capture — which TechCrunch reports is the apparent key to effective manipulation — flows into Genesis's training dataset.
The robots themselves wear the same sensor-equipped gloves, and their hands are designed to match human proportions. That fidelity, according to Xian, is what underpins capabilities like sorting and manipulating wires. Genesis has also invested heavily in simulation: adviser Minchen Li co-authored an influential physics algorithm that the startup uses to generate millions of high-fidelity simulated experiences for its model.
Eno, the flagship machine
Genesis's flagship humanoid is a robot called Eno, which TechCrunch likens to an iMac on wheels rather than a bare-metal skeleton, reflecting the feeling the company wants to convey to customers. Xian said the AI model performs better when running on Eno, but the robot was not ready to be demonstrated during the visit.
The company has built 30 to 40 Enos so far. It says the design of the mass-produced version has been finalized, a manufacturing contractor has been chosen to begin work next year in the US, and the machine is being piloted with a small number of undisclosed customers.
Hardware, not ideas, is the constraint
In Xian's assessment, the principal bottleneck is hardware — both mechanical design and the supply chains required for industrial-grade machines. Genesis's dexterous hands currently last only a few weeks of regular work before needing replacement, a limitation that sits awkwardly alongside any pitch for continuous operation.
On the software side, he is skeptical that architectural debates matter much. Whether the field talks about Vision-Language-Action models or World-Action models, he argues, everything ultimately reduces to transformers and attention, and he estimates the model architecture contributes perhaps 10% of the final result. What counts, in his view, is training infrastructure, data infrastructure and data quality. Nor does he expect another scientific breakthrough to be necessary: "we have all the recipes. It's just about execution," he told TechCrunch.
Why it matters
The round is one of the larger bets on physical AI, and it wagers that general-purpose robotic manipulation — long a stubborn unsolved problem — has become an engineering and data problem rather than a science problem. That thesis mirrors the argument that carried large language models forward: infrastructure and data quality over novel architectures.
The open question is reliability. A robot that packs boxes in a demo but fumbles them on a repeat run, with hands that wear out within weeks, is not yet deployable at industrial scale. With Figure, Agility and others chasing the same market, and investors wary of another robot winter if promises outrun performance, the $500 million gives Genesis the runway to close the gap between what its robots can show and what they can sustain.
- #robotics
- #humanoid-robots
- #startups
- #venture-capital
- #automation