Humyn Labs, a Bengaluru startup that records people doing real work so robots can learn from it, has raised $47 million in Series A financing, TechCrunch reported on November 18. The round combines $26 million in equity led by Accel with $21 million in venture debt led by Lighthouse Canton Growth Debt Fund II.
The company was founded earlier this year, and TechCrunch reports that about 90% of its revenue already comes from the U.S. For Bay Area robotics teams training humanoids, robot arms and world models, that number is the headline. The data these teams need is increasingly coming from vendors who film real workers around the world, and Humyn Labs is one of the fastest-growing names in that market.
Who invested in Humyn Labs' $47M Series A
According to TechCrunch, Accel led the equity portion alongside Cactus Partners, with participation from Prosus Ventures, BeeGlobal, Bajaj Finance's Finserv Intelligence and Bajaj Alts AI Opportunities Fund, Dallas Venture Capital, Lighthouse Canton and Jump Trading. Trifecta Capital joined the debt portion.
The debt has a practical purpose. TechCrunch reports that Humyn Labs pays its workers and data collection partners daily or weekly, while its AI lab customers typically pay 30 to 60 days later. That gap is a familiar cash flow problem for any data services company, and venture debt is one way to cover it without giving up more equity.
The founders are Manish Agarwal, the former CEO of Indian gaming company Nazara Technologies, and Ishank Gupta, a former executive at AB InBev. The two previously founded Kratos Studios, now called KGeN, whose contributor network helps Humyn Labs collect data in people's homes.
How Humyn Labs collects robot training data
Humyn Labs built its own two-camera hardware to record people doing everyday tasks, per TechCrunch. Collection runs across more than 20 countries in Latin America and Asia, with Brazil, India and South Korea as the main hubs. The company records people doing their usual jobs or household chores rather than staging activities, and its software tracks hand movements and how people interact with objects.
On its own site, Humyn Labs lists the inputs it captures: speech, first-person video, motion data from a 6-axis IMU sampled at 300 Hz, and grip force, along with stereo depth and hand keypoints. The site says its datasets cover 36 languages and target environments such as manufacturing, electronics assembly, logistics, retail, hospitality and agriculture.
Co-founder Ishank Gupta summed up the thesis in the company's announcement, quoted by PYMNTS: "A robot does not learn a job from the internet." TechCrunch also notes open questions about privacy and consent when recording inside homes in India, and the company's site lists consent audits among features marked as coming soon.
Traction, customers and the San Francisco push
Humyn Labs told TechCrunch it has 12 paying customers, including frontier AI labs, robotics companies and developers of world models, though it declined to name them. The company reported $10 million in revenue from April to October, $20 million in signed contracts and another $20 million not yet finalized. It has 51 full-time employees, most in Bengaluru, plus a network of local partners.
The new money goes toward deployment infrastructure in Brazil, Korea and India, more capture hardware and a bigger on-the-ground presence in San Francisco, London and Seoul, according to the release PYMNTS quoted. TechCrunch names Mecka AI, XDOF and MicroAGI as competitors in the same data race.
What it means for Bay Area robotics startups
Robotics companies have raised huge rounds this fall, from Genesis AI's $500M Series A to Agile Robots' Series F. A lot of that money ends up paying for data. Teams building models like the one behind Rhoda AI's video-trained robot need hours of real human motion, and few startups want to run camera crews in 20 countries themselves.
A few practical points for founders and operators weighing a vendor like this:
Ask about provenance and consent. Humyn Labs says its datasets carry verified provenance, and buyers should expect to document where training data came from as robots move into regulated workplaces.
Match the data to the task. Egocentric video of warehouse picking helps a picking robot far more than generic household footage, so specify environments up front.
Plan for sensor fit. Data with IMU and grip force readings is most useful when your robot's own sensors line up with what was recorded.
Budget for payment terms. If vendors carry 30 to 60 days of receivables, expect pricing to reflect it, and negotiate accordingly.
For hardware founders, this is also a reminder that the physical AI stack now has a full supply chain, from data vendors to model labs to prototyping spaces. Our guide to Circuit Launch in Mountain View covers where many local teams build their first robots.
The same week brought a big software-side AI infrastructure deal in the South Bay, which we cover in Palo Alto Networks' $4.5B Applied Compute acquisition.
Meet the South Bay robotics crowd in person
Tribe has been bringing the physical AI community together offline. On October 9, we hosted the Physical AI and Robotics Mixer at Circuit Launch in Mountain View. Tribe also organizes South Bay Renaissance to rebuild the South Bay and Peninsula startup and social scene one gathering at a time, with founder dinners, AI deep dives and demo nights. Follow the Tribe Events calendar on Luma to catch the next one.
Find your robotics people
If you're building robots or just love following the field, apply to join Futurists in Robotics, our Tribe for founders, engineers, operators and fans. If you do robotics research as a professor, PhD student or at a frontier lab, apply to join Embodied Frontier, the group built for researchers.