Nvidia backs startup Reactor as buzz grows for world models

World models are capturing the tech industry's attention and investment, as AI moves from screens to physical objects like cars, robots, and other devices. Several startups are developing these models, which are based on video data rather than text-based large language models. In the past week alone, AMD acquired World Labs for $8.2 billion, and Runway launched an open-weight version of its world model.

Reactor, a San Francisco-based startup, recently secured $74 million in funding from Nvidia's venture capital arm and Sapphire Ventures, bringing its total funding to that amount. Reactor's software accelerates and reduces the costs of running world models, which are sold as a cloud service. The company's co-founder and CEO, Alberto Taiuti, previously worked on Apple's Vision Pro and co-founded the AI video generator startup Luma AI.

Taiuti said world models could reduce the time and cost of building conventional simulations for robot testing, as they can generate environments from prompts. Reactor's technology can run world models that respond to robot actions, allowing developers to test their software in generated environments. This cloud-based approach gives developers access to more computing power without adding hardware to each robot and enables them to upgrade the computers running the models without replacing hardware across a fleet.

Reactor plans to use the new investment to secure more computing capacity, expand its team, and acquire robot hardware to test its systems.