Robots don’t fail on the demo. They fail on the dish that was put away wrong, the towel caught on a hinge, the guest who left a suitcase in the doorway. Maxinsights has raised $30 million to collect exactly that kind of messy footage.
The Santa Clara company closed a $30 million seed round led by Peak XV. Tola Capital, South Park Commons, INCE Capital and Alumni Ventures joined. The founding team and engineers come from Waymo, Google, Microsoft, Meta and Hulu.
A data company for physical AI
Maxinsights pays ordinary people to wear its camera gear while they do household chores, restaurant shifts and hotel housekeeping. The footage goes through a five-step pipeline: collect, check, annotate, curate and deliver. What comes out is egocentric video, shot from the worker’s point of view, ready to train robot models.
Its MaxLoop platform scores and fixes the capture as it comes in, so bad footage gets flagged before anyone labels it. A sensor glove, MaxGlove, is in development with Genesis AI to capture touch and force data. The two companies are aiming for a million hours of force data in 2027.
Large language models had the open web to learn from. Robot models have nothing like it. Every hour of useful data has to be recorded somewhere, by someone, with hands doing real things. That is the gap Maxinsights is selling into, and why investors call it a data layer.
The scale so far
The company says it has captured over 2 million hours, with 1.5 million ready for training. More than 5,000 contractors in six countries do the filming, hired through staffing firms, and the pipeline handles about 450,000 hours a month. It has offices in China and Indonesia, with Turkey and Mexico listed next.
Over 30 customers buy the data, eight on recurring contracts, including frontier AI labs and foundation model builders. Maxinsights supplied all 200,000 hours of egocentric data used to train Genesis AI’s GENE-26.5 model and is the largest data provider behind Dyna Robotics’ Dyna-2.
The founders
Chief executive Yuguang (YG) Yuan co-founded Mega Gear Robotics, an industrial robotics firm that was acquired, and before that ran product groups at Eaton. Co-founder and CTO Kun Wang was an early engineer at Google X and Waymo, then co-founded autonomous driving company QCraft, which grew past 1,000 employees and put its tech in more than a million vehicles.
Selling hours is not enough
Peak XV’s Rohit Agarwal put the problem plainly in the announcement: most companies in this space just sell hours. Raw volume is getting cheap. The value is in edge cases and in proving the data actually improves a model. Maxinsights is betting that its quality checks and its tie-ins with model builders like Genesis and Dyna make it more than a footage farm.
It joins a busy month for robot intelligence funding. Dream Labs raised $90 million to build shared robot brains. Those brains need data like this to learn from.