San Francisco-based infrastructure startup Reactor has secured $74 million in total funding to build the computational backbone for world models, a nascent class of AI designed to simulate physical environments rather than generate text. The latest capital injection includes new investment from Nvidia’s NVentures arm and Sapphire Ventures, expanding a Series A round initially announced in May.
The funding arrives as the artificial intelligence industry pivots toward spatial computing and embodied intelligence. While large language models dominate current enterprise applications, world models trained on video data are emerging as critical technology for robotics, autonomous vehicles, and interactive media. AMD’s recent $8.2 billion agreement to acquire World Labs and Runway’s release of open-weight world model architecture signal that major players view this sector as the next frontier beyond generative text and image synthesis.
Reactor addresses a specific bottleneck in this transition: inference efficiency. Unlike static content generation, world models intended for robotics require real-time responsiveness to dynamic inputs. Co-founder and CEO Alberto Taiuti noted that running these models at scale presents distinct latency and cost challenges that standard cloud GPU clusters are not optimized to handle. The company provides specialized software and managed compute services designed to reduce the operational expense of maintaining persistent, interactive simulations.
This infrastructure layer is particularly vital for robotic development. Traditional simulation requires labor-intensive manual construction of 3D assets and physics environments. World models can generate responsive testing grounds from natural language prompts or video references, potentially compressing development cycles. By hosting this computation in the cloud via Reactor’s platform, robotics companies can access high-fidelity simulation without embedding expensive inference hardware into every physical unit or maintaining proprietary server farms.
Nvidia’s participation extends beyond financial backing. The chipmaker has developed its own Cosmos world foundation models and Isaac robotics platform. Investing in Reactor secures Nvidia a position in the middleware stack necessary to make its hardware and foundational models commercially viable for third-party developers. This mirrors the ecosystem strategy employed during the LLM boom, where infrastructure providers became as valuable as the model creators themselves.
Reactor emerged from stealth in May with a founding team drawn from Apple’s Vision Pro division and Luma AI. The company reports active deployments with Hollywood studios and streaming platforms but identifies robotics as its primary growth vector. New capital will be allocated to securing additional compute capacity, acquiring physical robot hardware for validation testing, and expanding engineering headcount to support enterprise integration.





