Sanja Fidler, Nvidia's former vice president of AI research, has co-founded Veeda AI with two longtime research collaborators, Zan Gojcic and Huan Ling. The startup just landed $90 million in seed funding from Khosla Ventures and Radical Ventures. Veeda isn't building robots — it's building multimodal world models that generate virtual training environments for robots at scale.
Training robots by trial and error in the real world is expensive and slow. One mistake can damage the hardware, and you can't parallelize physical robots the way you can compute tasks. Veeda wants to move that trial-and-error into simulation, letting robots practice, fail, and retry in virtual environments. Fidler calls it the "Matrix" for physical AI.
These world models aren't just about generating realistic-looking video. They also need to accurately simulate space, motion, and object interactions — otherwise they're not reliable for training robots. The three founders spent years at Nvidia on related research, most recently working together on 3D and world simulation projects like Lyra 2.0 and OmniDreams. Fidler and Huan Ling also worked on NVIDIA's Cosmos world model platform.
Veeda is still early — it hasn't publicly shared specific models, customers, or a launch timeline. According to its website, the team is spread across Toronto, Zurich, Mountain View, and Singapore, and it's currently hiring for roles in world models, data, and large-scale training.
https://twitter.com/FidlerSanja/status/2090093857841267174