A year ago, Noitom Robotics' customers were asking for a thousand hours of recorded human movement. Now the same robotics teams are ordering hundreds of thousands of hours, according to founder and CEO Tristan Ruoli Dai. That shift in demand helped push the company to a valuation above $500 million and a No. 1 ranking in the Asia category of The Information's 50 most promising startups of 2026, published November 11 and detailed in a November 19 announcement.
Noitom Robotics spun off in 2025 from 13 years of motion-capture engineering, and runs what it calls data factories in Shenzhen and Mountain View, California, with data capture operations in Taipei and Kuala Lumpur. The company has raised more than $50 million in equity funding across three rounds since November 2025, most recently an oversubscribed Pre-A++ round in June 2026 backed by industrial funds including the Beijing AI Industry Investment Fund. It reports more than $10 million in revenue and over 100 employees.
Selling the raw material behind robot foundation models
Noitom Robotics sits in a crowded but fast-growing niche: companies that supply the human movement, manipulation and interaction data used to train robot foundation models, rather than building robots themselves. Its flagship product, ModalityNet, is what the company describes as an omnimodal data substrate, paired with dataset lines including HiPHI-MOV for whole-body motion and manipulation and HiPHI-OM for contact, force and tactile signals.
The company released its HiPHI dataset in August 2026, which it says has logged more than 15,000 downloads, and followed in November with HiPHI-Hand, a dataset focused on hand movement. Noitom Robotics says it now works with close to 100 companies across the physical AI field and is weighing expansion into Mexico and Indonesia to diversify where its human subjects are captured.
Competing in a data land grab
Noitom Robotics is not alone chasing this opportunity. Humyn Labs recently raised $47 million to capture human-activity video for robot training, and Mosaico and LanceDB have both raised rounds this year aimed at fixing data bottlenecks for physical AI developers. What separates Noitom Robotics, according to its own framing, is 13 years of prior motion-capture expertise rather than a startup built from scratch around the current embodied-AI boom.
The jump in order sizes that Dai describes — from thousand-hour requests to six-figure-hour requests — tracks with a broader industry pattern: humanoid and industrial robot makers are discovering that dexterous manipulation tasks require vastly more training examples than earlier robotics work assumed. Whether that demand sustains a $500 million valuation for a company with roughly $10 million in reported revenue will depend on how many of its nearly 100 client relationships convert into recurring, large-volume data contracts rather than one-off pilot purchases.