Genesis AI raises $500M to get Eno out of the lab and onto factory floors
$500 million for human-level robot hands
Genesis AI has raised a $500 million series A at a $3.5 billion post-money valuation. Premji Invest, Khosla Ventures and Eclipse co-led the round, which lands less than two years after the Paris and San Carlos-based company came out of stealth with a $105 million seed, one of the largest early-stage raises in French history.
That's almost 5X the seed, in one round.
The plan for the money is fairly simple: more human data, more training, harder manipulation tasks, more hiring and, most importantly, Eno robots working at customer sites instead of in Genesis's own office. Industrial customers come first, in manufacturing, logistics and laboratories. Hospitality and healthcare are next, followed by consumer and outdoor applications.
“Our mission is to remove the physical limits on human ambition,” CEO and co-founder Zhou Xian said in a statement. “To accomplish this, we’re working to build the world's best general-purpose robot, one that works reliably in the real world, day after day. That requires the intelligence, hardware, and everything that goes into training and testing the robots to work as one system.”
Genesis AI has the full stack: foundation model, hardware, data collection, simulator
The key phrase there is "one system."
Genesis builds the whole stack itself. There's GENE, the robotics foundation model. There's a dexterous robot hand with 20 actuated degrees of freedom, plus a data collection glove built to match it, so that a human hand, the glove and the robot hand form what Genesis calls a 1:1:1 mapping. People wear the gloves while doing their regular jobs, and the data trains the robot directly. Then there's Genesis World, a simulator for training and testing skills virtually. And finally there's Eno, the robot that puts it all to work and sends deployment data back into the loop.
The company's thesis is that as robots move from single tasks to months of autonomous operation across whole workflows, the system is only as reliable as its weakest part. Building everything in-house lets Genesis optimize every layer together and avoid supplier constraints.
It's the Apple playbook, basically. It's also expensive, which goes some way toward explaining why a company with no publicly named customers needs half a billion dollars.
“General-purpose robotics is a systems problem, and Genesis is attacking the whole system: a data engine built for massive scale, world-class simulation, human-like dexterity and Eno’s remarkably beautiful form factor,” says Eclipse partner Charly Mwangi, who also backed the seed round.
Eno, the "remarkably beautiful form factor"
Ah yes, the form factor.
Eno is a fair test case for a site that tracks humanoid robots, because Eno isn't trying very hard to be one. It has no head and no face. It's a tower of articulated panels on a wheeled base with two arms and two very capable hands. Genesis describes it as "human in function, not in form." When I saw it in San Carlos in June, I called it the iPhone moment for humanoid robots, half seriously: it felt more like a 10th-generation product than a first one.
Genesis's bet is that the hands are what matter, and the rest of the body mostly exists to put them where they need to be. Plenty of roboticists agree. Back in May, Genesis showed GENE-26.5 cooking a 20-step breakfast: cracking eggs, chopping tomatoes, taping wires and solving a Rubik's cube, autonomously and at 1X speed.
(I want a robot that makes me breakfast. Badly.)
“Most human physical skills have never been recorded, so we use gloves that capture movement and touch as people work, and we’re designing one of the most dexterous robot hands to match ours exactly,” co-founder Théophile Gervet says. “Together, they let Gene, our foundation model, learn directly from people and approach human-level dexterity.”
Genesis is far from alone in thinking human glove data is the shortcut to robot dexterity. Reward AI says its OM-1 model was trained only on glove data, with no teleoperation, and now runs humanoids doing last-mile delivery. Another startup I covered last month, which is building a glove-based data engine, is testing its MaxGlove with Genesis. And touch is getting its own models and its own money: Ant Group's Robbyant just open-sourced LingBot-Touch, and Zurich-based Enact Intelligence recently came out of stealth with $30 million for robot touch.
The investors in Genesis AI's series A
The investor list is long.
Beyond the three co-leads, Durable Capital Partners, accounts advised by T. Rowe Price, Bpifrance, HSG, NVIDIA, AMD Ventures, Zoom, Samsung Next and Ericsson joined, along with former Google CEO Eric Schmidt, Zoom CEO Eric Yuan, former Yahoo CEO Jerry Yang, French entrepreneur Xavier Niel and former Tesla CFO Zach Kirkhorn.
“I've believed in this team's vision since the beginning, and what they've accomplished in such a short time is extraordinary,” Schmidt said in a statement.
Premji Invest's Vedant Agrawal was more specific about what convinced him.
“What stood out to us was watching Genesis robots complete long, multi-step tasks on their own at the same speed a person would, with smooth and natural movement,” he said. “Completing full tasks in real time, not just short clips, is what turns a robot into something a business can rely on.”
That matters, because robotics has a demo problem. Sped-up video, cherry-picked takes, a teleoperator just out of frame: we've all seen it. When I wrote about Genesis's hands in May, Kyber Labs co-founder Tyler Habowski summed up the state of the art for me.
“There are literally zero robot hands deployed right now doing routine work,” he said. “The best hands are hundreds of thousands of dollars, and they break all the time.”
What's needed next? Customers
Genesis is betting $500 million it can be the exception.
But the announcement leaves out a lot of what a buyer would want to know. No customers are named. There are no shipment numbers, no price and no durability data. When I saw Eno in June, payload was roughly 3-5 kilograms per arm and battery life was around four to six hours, both still being optimized. Those numbers will need to move for a lot of factory and warehouse work.
The competition isn't waiting, either.
HumanoidMegaHub tracks about 600 humanoid robots from more than 400 manufacturers, including Tesla, Figure, Agility and 1X in the U.S., plus Unitree, AgiBot and UBTech in China. Some are starting to post real deployment numbers: Apptronik, for instance, recently signed a deal to place up to 1,000 Apollo robots in GXO logistics facilities.
Lots of companies in this space have big money and slick demos. Far fewer have robots doing real work, every day, for paying customers.
Genesis now has the money. Next I'm hoping to see some customer names.
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.”