Build AI is a San Francisco-based research lab and public benefit corporation that produces training data for Physical AI. It describes itself as a data hyperscaler, vertically integrated across hardware design, manufacturing, logistics, field data collection, annotation, infrastructure and model training. The company's stated mission is to solve physical labour and unlock human potential. It has raised more than $30 million and employs fewer than 30 people, with operations in San Francisco and Shenzhen.
The technical work spans the full stack. Engineers work on custom wearable data-collection hardware designed in Shenzhen, large-scale data pipelines, annotation systems, customer-facing APIs and model-training infrastructure. Research areas include video and world-model research, embodied AI, robotics and 3D pose estimation. The company's Egocentric series captures real-world physical labour as monocular 1920×1080 video at 30fps from workers in industrial environments, and Build AI supports an open training set for the research community.
Build AI has set a research roadmap target of 100m hours of data. It offers service-level guarantees covering 3D-pose accuracy, diversity and quality, and partners with research laboratories working on video, world models, embodied AI and robotics. The company was founded by Patrick Rim, Eddy Xu, Jonathan Jia and Zikang Jiang.
