According to data from PitchBook cited by the Financial Times, robotics and physics AI companies have raised nearly $48 billion since the beginning of 2026, with an increasingly large portion of this funding flowing into an input the industry still lacks: real-world video of humans performing everyday tasks.
The FT reports that labs including PhysicalIntelligence, SkildAI, NEURARobotics, and Generalist are combining external data providers with internal data collection to feed their models. Teams are collecting video in "robotic gyms"—where workers operate robotic arms to complete tasks;
simultaneously, they are sending recording equipment into homes, offices, and factories to film people's daily work. In their words, the goal is to capture the movement, contact, and force feedback in scenarios where robots will have to handle them independently in the future.
Unlike text and images, video of everyday operations has never been crawled from the public internet; it can only be custom-collected and annotated frame by frame. NEURA has described high-quality real-world training data as "the scarcest resource for physics AI" and is collaborating with partners including RWTH Aachen University and the Technical University of Munich to build the NEURAGym network, with plans to have five sites operational in Europe, the US, and China by the end of 2026.
This funding round continues the expansion we've seen throughout the year for our robotic trackers. The FT believes that the key constraint currently hindering the industry's development is not hardware or models, but the data gap.
