Robots are only as capable as the data they train on — and Maxinsights wants to supply it. The Santa Clara-based startup has raised $30 million in seed funding led by Peak XV, with participation from Tola Capital, South Park Commons, INCE Capital and Alumni Ventures, and launched two new tools aimed at making real-world training data for physical AI cheaper and better.
The company collects first-person footage of people doing everyday work — loading dishwashers, preparing food, washing cars — captured by camera hardware worn by more than 5,000 contractors across six countries. Maxinsights says it has captured over 2 million hours of footage, with 1.5 million hours model-ready, and counts more than 30 customers, including AI labs and foundation-model developers, with eight on recurring contracts.
Its data has already powered notable models: all 200,000 hours of egocentric data behind Genesis AI’s GENE-26.5 came from Maxinsights, which also describes itself as the largest data provider for Dyna Robotics’ Dyna-2 World-Action Model. In the open-source world, datasets shared through platforms like Hugging Face play a similar role for researchers, while Databricks and similar platforms handle the data engineering pipelines such operations require.
The two new products attack the cost problem from both ends. MaxLoop is software that scores incoming footage and gives operators same-day feedback, replacing a one-to-two-week review cycle and cutting review costs by about 80%. MaxGlove, developed with Genesis AI, is a sensor glove capturing grip force, pressure and slip signals that video alone cannot provide, with a goal of one million hours of force data in 2027.
The startup was founded by Yuguang (YG) Yuan, previously of industrial robotics company Mega Gear Robotics, and CTO Kun Wang, a Google X and Waymo alumnus who later co-founded autonomous-driving firm QCraft.
Source: The AI Insider
