Physicl Raises $10 Million From Inovia to Sell Robot Builders a More Faithful Digital World: 10,000 3D Environments, 100,000 Assets

Alex de Vigan spent a decade building nfinite, a platform that let brands and retailers render photorealistic 3D product imagery for e-commerce. Generative AI then swamped that business — so he is betting the same technology stack can solve a harder problem: making digital worlds faithful enough to train robots. His new venture, Physicl, has raised $10 million in seed funding from Inovia Capital.

The problem Physicl targets is well known to anyone training physical AI models. Video data delivers poor returns because it is passive and uncontrolled; teleoperated real-robot data is expensive to collect at scale; and simulations are cheap but only useful if their geometry, materials and physics actually match reality. "What we were super unique about doing in 3D, AI was going to replace," de Vigan told TechCrunch, "but something was missing in AI: its ability to understand, interact, and manipulate space and objects."

The company's first product, Studio, launches with 10,000 3D environments and 100,000 assets representing various objects. Users build scenes to generate synthetic training data for specific failure modes — a robot that cannot find a cup on a cluttered countertop, or cannot recognize it in shadow. The pitch is post-training on hard edge cases rather than bulk pretraining data.

Physicl already has a credibility signal: SpAItial, a world-model builder, used it as an early customer before releasing its latest product, Echo-3, last week. SpAItial founder Matthias Niessner, also a robotics professor at the Technical University of Munich, said his team needed exactly this kind of high-fidelity synthetic data. "The real challenge is, you have to curate the data by hand, and this is a very long cycle," Niessner told TechCrunch. "You're running Claude, and it checks, 'Oh, I need more data of this in my training,' and then it auto-generates that data depending on the training state. That is exactly what Physicl is offering."

De Vigan is candid that the category is crowded and unclear — "the space is very young and very blurry sometimes," he admitted — with many startups collecting, labeling or generating data in different modalities for the same customers. His longer-term ambition is to be tooling rather than a data vendor: customers feeding in trillions of images or millions of hours of LiDAR captures, which the platform converts into high-quality 3D representations and then into billions of training permutations.

It is a crowded field, and $10 million is a modest war chest next to the nine-figure rounds flowing into robotics foundation-model companies. But the thesis is aimed at a real bottleneck: the gap between what simulations show and what physics does. Whoever closes that gap cheaply becomes infrastructure for the whole physical AI stack — which is precisely the position de Vigan says he spent ten years accidentally building toward.