Genesis AI raises $105 million seed from Eclipse and Khosla to build a universal robot brain on its own simulation engine

Genesis AI, a robotics lab with offices in Paris and Palo Alto, came out of stealth on July 1, 2025 with $105 million to build what it calls a universal robotics foundation model and a horizontal robotics platform. Eclipse and Khosla Ventures co-led the financing, which the company later described as seed funding. The announcement went out at 09:00 Eastern time and did not state a valuation.

The round

Besides the two co-leads, the release named Bpifrance, the French state investment bank, the investment firm HSG, former Google chief executive Eric Schmidt and French telecoms entrepreneur Xavier Niel as backers. In a March 11, 2026 release the company again put the total at $105 million in seed funding and added AI researchers Daniela Rus and Vladlen Koltun to its list of supporters. A seed round of that size was unusual for a company that had not yet shown a product, a customer or a robot of its own.

Genesis presented itself as a global physical AI research lab and a full-stack robotics company. Its stated mission is to "unlock unlimited physical labor" by putting general-purpose robots to work on the physical tasks that underpin the economy. The release framed the market with four figures of its own. Physical labour accounts for $30 trillion to $40 trillion of global GDP, more than 95% of it is still not automated, 75% of companies worldwide struggle to fill jobs, and in most automated industries the ratio of robots to human workers rarely goes above one to 30.

A data engine before a model

The core of the pitch is data, not hardware. Genesis said it is building a scalable data engine that combines high-fidelity physics simulation, multimodal generative modelling and large-scale collection of real robot data. The simulation stack was developed in house, according to the company, and is meant to produce synthetic training data at a scale that teleoperation alone cannot reach. Kanu Gulati of Khosla Ventures described the result as a continuous closed loop in which simulation data and real-world robot data feed each other.

In plain terms, a robot foundation model learns to map what a robot sees and is told to the motions it should make, and it needs millions of examples to generalise. Collecting those examples with human operators steering real arms is slow and expensive. A fast physics simulator can generate many more attempts, including failures, at almost no marginal cost, but policies trained only in simulation often break when they meet real friction, lighting and clutter. Genesis is betting that mixing both sources in one pipeline narrows that gap faster than either alone. The company also said it plans to open-source components of its data engine and of its foundation model.

The team already had a public track record in simulation. Genesis World, an open-source physics platform for robot learning, began as an academic project in December 2024 and is now developed with support from Genesis AI, according to its documentation. The code is released under the Apache 2.0 licence. Its documentation today describes a unified multi-physics engine, a photorealistic renderer called Nyx and a cross-platform compiler called Quadrants behind a single Python interface, and the GitHub repository had about 30,000 stars by October 2026.

Who is behind it

Zhou Xian leads the company as chief executive, and the March 2026 release names Théophile Gervet as his co-founder. According to the launch release, the founding team came from Mistral AI, Nvidia, Google, Carnegie Mellon University, MIT, Stanford, Columbia and the University of Maryland, with backgrounds in physics simulation, graphics, robotics and large-scale model training. The operating company is a French SAS, which gives Europe a rare home-grown entrant among well-funded robot brain developers.

Zhou said in the release that general-purpose robots powered by physical AI would define the next major chapter of human history. Charly Mwangi, a partner at Eclipse, said general-purpose robotics stands to affect trillions of dollars of labour value across sectors. In its own post on the deal, Eclipse tied the investment to labour shortages in manufacturing, construction, healthcare and agriculture.

How it compares

The round put Genesis among the better-funded companies chasing a general robot brain, though well short of the largest. Skild AI raised a $300 million Series A at a $1.5 billion valuation in July 2024. Dyna Robotics announced a $23.5 million seed round in March 2025. On the humanoid side, Figure AI raised $675 million in a Series B in February 2024 and 1X Technologies $100 million in a Series B in January 2024. By a MEKAWIRE calculation, Genesis's seed was 4.5 times Dyna's seed and 35% of Skild's Series A.

The difference is in what each company sells. Figure and 1X build their own humanoids and train models for them. Skild AI and Physical Intelligence offer models meant to run on many robot bodies, and NVIDIA supplies simulation, models and onboard computers to almost all of them. Genesis is closest to the model companies but claims to own the simulator as well, which would let it control the cost of its main input.

Risks and open questions

At launch Genesis named no customers, no robot hardware partners, no model release date and no valuation. Its claims about the simulator's role rest on the company's own description, and the gap between simulated and real-world performance has humbled many earlier efforts. The promise to open-source parts of the stack also leaves open how much of the value Genesis can keep once rivals can use the same tools.

What came next

By March 2026 the company had hired Vivian Sun, previously head of automated driving at Amazon, as vice president of commercial and strategy, and that release carried a Paris and San Carlos, California dateline. Its website now promotes a product called Eno with targeted customer deployments planned by the end of 2026.