Veeda AI raises $90M seed to build world models for robot training

Veeda AI raises $90M seed to build world models for robot training

Toronto-based Veeda AI has raised more than $90 million in a seed round backed by Radical Ventures and Khosla Ventures, according to The Logic. The company was founded by Sanja Fidler, Zan Gojcic and Huan Ling, all former Nvidia AI researchers, with Fidler serving as CEO. Veeda is building multimodal world models trained on sensor and physical-world data to generate simulated environments where robots can learn through trial and error, rather than relying primarily on imitation learning or slow, costly, and potentially unsafe real-world testing.

The bet targets a real bottleneck in physical AI: robots trained mainly by imitating human or robot demonstrations struggle to generalize past what they've seen, and real-world trial-and-error is too expensive and risky to run at scale. Fidler has described Veeda's goal as building a virtual proving ground where a robot's AI can encounter far more variation in objects, tasks, and conditions than physical testing allows. That frames world models as an infrastructure layer sitting underneath robotics programs rather than a robot product itself — a layer that several well-capitalized teams are now racing to establish before hardware makers standardize on a particular training stack.

For robotics builders, Veeda adds another simulation-layer option to weigh against physics-engine incumbents and rival world-model efforts, with the founders' Nvidia pedigree likely to shape early hardware and chip partnerships. For investors, a $90 million seed for a pre-product infrastructure company is a large bet this early, signaling that backers see the simulation layer as valuable and defensible enough to fund well ahead of commercial deployment or revenue.

#VeedaAI #WorldModels #Robotics #PhysicalAI #SeedFunding #Nvidia