The artificial intelligence landscape is witnessing a pivotal shift towards embodied AI, where intelligent agents need to operate and learn within physical environments. However, training these systems in the real world is often unsafe, costly, and difficult to scale. Veeda AI addresses this challenge by developing "world models" – multimodal foundation models that simulate physical reality, allowing AI agents to learn through interaction in infinitely scalable virtual environments.

Founded by former Nvidia researcher Sanja Fidler, along with Huan Ling and Zan Gojcic, Veeda AI has secured over $90 million in seed funding. This round, co-led by prominent AI investors Khosla Ventures and Radical Ventures, represents one of the largest seed financings ever for a Canadian startup. The capital will primarily be used to expand the research team and acquire the necessary compute resources to train these complex generative world models.

Veeda AI's approach aims to provide high-fidelity simulated environments, akin to building "the Matrix" for physical AI, where robots can safely make mistakes and learn without real-world limitations. This contrasts with traditional methods that rely heavily on real-world trial and error or less sophisticated simulations. By abstracting the physical world into a digital twin, Veeda AI seeks to accelerate the development of robust and adaptable physical AI systems, which are crucial for advancements in robotics and autonomous systems.

The company's focus on world models is a strategic bet on how embodied intelligence will ultimately scale. While large language models have transformed digital interactions, the physical world presents unique complexities. Veeda AI's platform provides a crucial layer for developing AI that can understand physics, spatial reasoning, and temporal coherence, enabling more capable and versatile robots.