Maxinsights

Maxinsights (legal name Maxinsights Corporation, styled MaxInsights) is a San Francisco Bay Area data company that supplies training and evaluation data for physical AI and robot foundation models. It provides large-scale multimodal data collection, hand tracking, upper-body and whole-body motion capture, tactile force signals, simulation and data evaluation for generalist robotics, vision-language-action (VLA) models and world action models (WAM).

The company's core offering is egocentric (first-person) video of people performing real-world tasks, annotated with additional signals. Maxinsights lists nine output measures that can accompany its data: video understanding, depth, object tracking, hand poses, upper-body poses, tactile force, localization, segmentation and point clouds. Tasks are graded by action complexity (easy, medium and hard). The company states cumulative collection of 2.0 million hours of data, production capacity of 450,000 hours per month and coverage of more than 50,000 scene types, and presents a "10M" hour frontier as its scale target. Its stated strategy is to scale both the volume of recorded human experience and the density of physical information captured per hour. Beyond collection and annotation, it offers model evaluation services for robotics teams.

The company states that its data supports generalist robotics and world-model work at OpenAI, Google DeepMind, Meta, Figure, 1X, Skild, Genesis AI and Dyna Robotics. In September 2026 the company published "Beyond 1M Hours: The Data Scaling Laws of Physical AI", an article discussing how models such as Dyna Robotics' Dyna-2 (pre-trained on more than one million hours of human egocentric video) and Genesis AI's models combine egocentric video, third-person video and instrumented gloves for hand motion and tactile signals, and arguing for richer state representations that include hand pose, body configuration, object geometry, contact signals and 3D environment data.

Maxinsights was founded in 2024 by Kun Wang and Yuguang Yuan, is headquartered in San Francisco, California, and has 11-50 employees. It has raised two seed rounds, including one dated 25 November 2025, from investors including HSG, Alumni Ventures, Top Harvest Capital, South Park Commons, Tola Capital and Plug and Play. It operates as a human-centric data infrastructure platform for physical AI and embodied robotics and can be contacted at [email protected].