Humyn Labs Raises $47M and Books $10M in Revenue Selling Video to Train Robots
Teaching a robot to wash dishes takes a lot of first-person video with labeled hands and objects, and the web has almost none of it. India's Humyn Labs has raised $47M in a Series A: $26M in equity led by Accel and $21M in venture debt. The startup films people in 20+ countries with its own dual-camera device and sells the labeled footage. Revenue from April through October was $10M. For roboticists, that is a sign this kind of data can already be bought.
AI-processed from TechCrunch; edited by Hamidun News
Language models learned from the internet, where text is abundant. Robots are a different story: the web has almost no video of a person washing dishes, unpacking groceries or folding laundry from a first-person view, let alone with labels showing what the hands are doing. Humyn Labs, based in Bengaluru, India, is filling that gap.
It has raised $47M in a Series A (the first major institutional round after seed funding). Of that, $26M is equity, and $21M is venture debt: a loan to a startup that is repaid in cash rather than with a stake in the company. The company also reported $10M in revenue from April through October.
That is how much it has already been paid for video that shows robots how people do household tasks. For anyone building robotics or world models (AI that learns to predict how the physical environment works), this is a signal: data is turning into a separate business, with its own money and its own problems.
Who put up the money, and how
The first $26M is ordinary equity, led by Accel. Prosus Ventures, BeeGlobal, Finserv Intelligence and Bajaj Alts AI Opportunities Fund (entities of the Bajaj group), Dallas Venture Capital, Lighthouse Canton and Jump Trading joined the round. The other $21M is venture debt, led by Lighthouse Canton Growth Debt Fund II with participation from Trifecta Capital.
The company was founded in February. Its founders are Manish Agarwal, a former CEO of the gaming company Nazara Technologies, and Ishank Gupta, who previously held a senior role at the brewing giant AB InBev. Before this, they built Kratos Studios together, now known as KGeN. KGeN has a network of contractors in developing countries, and that network helps Humyn Labs film people in their homes.
How the data collection works
The setup is down to earth. Humyn Labs built its own device with two cameras, which is used to record people performing ordinary tasks. Software then processes and labels the video: it marks what is happening in each segment and tracks hand movements and interactions with objects. The raw footage becomes a dataset that can be used to train a model.
Filming takes place in more than 20 countries in Latin America and Asia, with Brazil, India and South Korea as the main hubs. The founders stress that they film people doing their regular work or household chores, not staged scenes. Local partners select the sites and oversee the process. According to Agarwal, contractors get an incentive, but they have to be supervised so that the output is the right data. The company's central thesis is that scale, diversity and quality must grow at the same time.
The business numbers and why it needs debt
The startup has 12 paying customers: frontier AI labs, robotics companies and world-model developers. About 90% of revenue comes from the US. The founders did not disclose customer names. The company earned $10M from April through October, another $20M is locked in by signed contracts, and a further $20M in deals is not yet closed.
The venture debt here is not a whim. Contractors and collection partners are paid daily or weekly, while money from AI labs arrives after 30–60 days. The loan covers that cash gap and dilutes the founders less than issuing additional shares would. The company has 51 employees, most of them in Bengaluru, plus a network of local partners.
Competitors and honest limitations
Humyn Labs is not alone. Mecka AI, XDOF and MicroAGI collect similar data, and in recent months India has become a hub for gathering training data for physical AI: Human Archive and BuildAI also operate there, drawing on the large labor market.
This model has a weak spot: privacy and consent. When people are filmed at home, questions arise about what exactly ends up in the frame, who later sees the recordings and whether the participant understands where the footage will go. Agarwal answers with oversight through local partners, but the article contains no details on consent rules or data protection. The revenue and contract figures were also stated by the founders themselves and have not been independently verified.
What this means for practitioners
First: if you are building a robot or a world model, the bottleneck is often not the architecture but first-person video with well-labeled hands and objects. Buying such data from contractors is already realistic, and the supplier market is growing.
Second: for founders, this is an example of a business with heavy working capital needs. Contractors have to be paid fast, while customers pay after two months, so debt becomes a working tool.
Third: when choosing a vendor, ask not only about the volume of video hours but also about how the people in the frame gave consent. For a product headed to heavily regulated markets, this can turn into a legal risk.
- Nov 20, 2026Four Hours on an RTX 5090 Cost 348 ₽, or From 1 730 ₽ With a One-Day Minimum
- Nov 20, 2026In China, Job Postings for AI-Exposed Professions Fell 10 Percentage Points Faster Than the Rest
- Nov 20, 2026Suno adds Visuals: turn a song into a clip of up to 15 seconds with lip-sync, for separate credits
- Nov 20, 2026HeyGen claims 84% of viewers mistook its Avatar 6 AI avatar for a real person
Key Questions Answered
Who put up the money, and how?
The first $26M is ordinary equity, led by Accel. Prosus Ventures, BeeGlobal, Finserv Intelligence and Bajaj Alts AI Opportunities Fund (entities of the Bajaj group), Dallas Venture Capital, Lighthouse Canton and Jump Trading joined the round. The other $21M is venture debt, led by Lighthouse Canton Growth Debt Fund II with participation from Trifecta Capital. The company was founded in February.
How the data collection works?
The setup is down to earth. Humyn Labs built its own device with two cameras, which is used to record people performing ordinary tasks. Software then processes and labels the video: it marks what is happening in each segment and tracks hand movements and interactions with objects. The raw footage becomes a dataset that can be used to train a model. Filming takes place in more than 20 countries in Latin America and Asia, with Brazil, India and South Korea as the main hubs.
What are the key details on The business numbers and why it needs debt?
The startup has 12 paying customers: frontier AI labs, robotics companies and world-model developers. About 90% of revenue comes from the US. The founders did not disclose customer names. The company earned $10M from April through October, another $20M is locked in by signed contracts, and a further $20M in deals is not yet closed. The venture debt here is not a whim. Contractors and collection partners are paid daily or weekly, while money from AI labs arrives after 30–60 days.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.