On November 18, 2026, Bengaluru-based Humyn Labs announced it had raised $47 million in Series A funding to expand its collection and processing of real-world data for robotics and physical AI developers. The round is split into $26 million in equity led by Accel and $21 million in venture debt led by Lighthouse Canton, and it arrives less than a year after the company was founded in February 2026. Humyn Labs collects footage of people performing everyday tasks across more than 20 countries, using proprietary two-camera recording devices and automated vision labeling software, and sells the resulting datasets to the AI labs and robotics companies that need real-world data to train machines that operate in physical environments. The company says it has 12 paying customers, reported $10 million in revenue between April and October, and claims to have surpassed $50 million in annualized revenue within six months of starting operations.
The fundraise is a bet on a bottleneck. The AI industry has spent the last several years solving the problem of training models on text, images, and video scraped from the internet, and that approach has produced remarkable systems. But robots cannot learn to fold laundry, load a dishwasher, or assemble a product from internet data alone, because the internet contains very little footage of the fine motor activity involved, and almost none of it labeled in the way a training run requires. Physical AI, the field of artificial intelligence that enables machines to understand and perform complex actions in the real world, needs data that does not exist yet, and someone has to go out and collect it.
The round at a glance
The $47 million Series A is structured as two separate financings announced together. The equity portion is $26 million, led by Accel, with participation from Prosus Ventures, BeeGlobal, Finserv Intelligence through Bajaj Finance, the Bajaj Alts AI Opportunities Fund, Dallas Venture Capital, Lighthouse Canton, and Jump Trading, alongside angel investor Anita Pramoda, the former chief financial officer of Epic. The venture debt portion is $21 million, led by Lighthouse Canton's Growth Debt Fund II, with participation from Trifecta Capital. The debt facility is designed to address working capital needs between immediate payouts to data contributors and delayed remittances from clients, a timing gap that is structural to a business that pays people to record video and gets paid by customers on longer cycles.
The split between equity and debt is informative. A company that raises $26 million in equity and $21 million in debt in the same round is signaling that it has predictable, contract-backed cash flows that a lender is willing to fund, and that it would rather use debt for working capital than dilute its shareholders further. Venture debt of this size, led by a named fund with a specific mandate, is not a rescue facility. It is a working capital tool for a business whose constraint is timing, not demand. For founders, the structure is worth studying: Humyn Labs is using each kind of capital for the job it does best, equity for building the platform and the collection network, debt for bridging the gap between paying contributors and collecting from customers.
The investor list is also notable for its breadth. Accel, one of the most established venture firms in India and globally, leads the equity. Prosus Ventures, the investment arm of one of the world's largest technology investors, joins alongside. Bajaj Finance's investment arms bring the backing of one of India's largest non-bank lenders, and Jump Trading brings a quantitative trading firm's appetite for data businesses. Dallas Venture Capital and BeeGlobal add further institutional weight. That a nine-month-old company assembled a syndicate of this quality, and did so while reporting revenue, is a measure of how seriously investors now take the physical AI data problem.
What Humyn Labs does
Humyn Labs describes itself as a research engine and data platform for physical AI, and the description is precise about what it sells. Its customers are frontier AI labs, robotics companies, and developers of world models, the systems that learn to predict how the physical world behaves. Those customers need training data that shows how humans actually move through the world: how a hand grips a mug, how a body navigates a crowded kitchen, how an object is picked up, carried, and set down. That data is not available on the internet in usable form, because the internet was not built to record it, and the data that does exist is rarely labeled with the precision a training run demands.
Humyn Labs solves the problem by going into the real world and recording it. The company operates across more than 20 countries, with primary hubs in India, Brazil, and South Korea, and it captures authentic human tasks in unscripted environments, from homes to workplaces to public spaces. Its field teams use proprietary two-camera recording devices to capture the same activity from multiple angles, and automated vision labeling software to annotate the footage: identifying hands, objects, movements, and interactions at a scale that manual labeling could never match. The output is commercial proprietary video datasets and annotation services, sold to customers who use them to train models that must operate in physical environments. The company also publishes BRIDGE, an independent benchmark series for physical AI covering voice, vision, and touch, which positions it as an authority on what good training data looks like.
The founders
Humyn Labs was founded in February 2026 by Manish Agarwal and Ishank Gupta, two serial entrepreneurs whose backgrounds explain the company's unusual combination of consumer instinct and operational discipline. Agarwal is the former chief executive officer of Nazara Technologies, the Indian gaming company, and an alumnus of the Indian Institute of Management Ahmedabad. Gupta is a former executive at brewing conglomerate AB InBev and an alumnus of the Wharton School. The two previously worked together on gaming and digital community businesses, and they bring to Humyn Labs a shared understanding of how to build and scale a consumer-facing operation across India and beyond.
That background matters because data collection at Humyn Labs' scale is, at its core, a people operation. The company must recruit, train, equip, and pay contributors in more than 20 countries, maintain quality standards across languages and cultures, and move footage through a processing pipeline fast enough to be useful to customers who are training models now. That is an operations problem as much as a technology problem, and it rewards founders who have run large, distributed organizations before. Agarwal's experience leading a listed gaming company and Gupta's experience inside one of the world's largest consumer goods companies are directly relevant to the business of coordinating thousands of people to produce a standardized product.
The speed of the company's progress also reflects its founders' networks. Raising a $47 million Series A within nine months of founding, from a syndicate that includes Accel and Prosus, requires relationships and credibility that first-time founders rarely possess. The founders' track records gave investors a reason to believe that a team formed in February could be reporting $10 million in revenue by October. For founders watching from the outside, the lesson is not that speed is easy, but that it is compounding: a strong operating history buys the trust that makes a fast start possible, and a fast start buys the momentum that makes the next round easier.
The data bottleneck in physical AI
The core insight behind Humyn Labs is that physical AI has a data problem that is structurally different from the data problem of language models. Language models were trained on the internet, which contains trillions of words produced by humans for other humans. The data was already there, waiting to be collected. Physical AI has no equivalent corpus. There is no internet of human movement, no archive of how hands manipulate objects, no library of how bodies navigate cluttered rooms. The data that robots need to learn from must be created deliberately, by recording people doing real things in real places, and labeling what happened with enough precision for a model to learn from it.
That gap is now the binding constraint on robotics. The models themselves have improved dramatically: vision-language models can describe a scene, plan a sequence of actions, and reason about objects in ways that were impossible a few years ago. But a model that can describe how to fold a shirt still cannot fold one, because it has never seen enough examples of folding, labeled at the level of finger positions and fabric tension, to learn the motor skill. The industry calls this the data bottleneck, and it is the reason that robotics companies are now spending heavily on data collection, simulation, and teleoperation. Humyn Labs is betting that real-world recorded data, collected at scale and labeled well, is the fastest path through the bottleneck.
The market opportunity is correspondingly large. A July 2026 report from Counterpoint Research estimated that India's data collection and annotation services for physical AI could unlock a $10 billion opportunity by 2030, a figure that reflects both the scale of the demand and India's structural advantages in supplying it: a large, young workforce, low collection costs relative to the United States and Europe, and a deep pool of engineering talent to build the processing pipeline. Humyn Labs, based in Bengaluru with collection hubs across Latin America and Asia, is positioned directly in the middle of that opportunity, collecting data where it is cheapest to collect and selling it where it is most valuable, with roughly 90 percent of its revenue originating from United States developers.
The two-camera rig and the collection network
The technical heart of Humyn Labs is its capture system: proprietary two-camera recording devices paired with automated vision labeling software. The two-camera design matters because a single camera cannot see everything that matters in a manipulation task. A hand gripping an object occludes the object from one angle; a body turning away from the camera hides the movement of the other side. Two cameras, recording the same activity from different angles, produce a richer record, and the combination lets the labeling software reconstruct what happened in three dimensions rather than guessing from a flat image. The company says the system is designed to capture fine motor activity, object manipulation, and natural movements in unscripted environments, which is a precise description of the hardest data to collect and the most valuable to have.
The collection network is the other half of the system. Humyn Labs operates across more than 20 countries throughout Latin America and Asia, with its main hubs in India, Brazil, and South Korea, and it plans to use the new funding to build deployment infrastructure in those three countries, strengthen its data capture hardware and processing engine, and expand its on-the-ground presence in San Francisco, London, and Seoul. That footprint is deliberate: India and Brazil offer large pools of contributors and diverse environments, South Korea offers a technologically advanced market with a strong robotics ecosystem, and San Francisco, London, and Seoul put the company close to its customers. The funding will support all three of those priorities at once, which is why the round is as large as it is.
Scale is the point. A dataset of a few thousand hours of footage is a research project. A dataset of millions of hours, collected continuously across dozens of countries and labeled consistently, is an asset that compounds: every new customer benefits from the data collected for the last one, and every new hour of footage makes the existing corpus more valuable. That compounding is what investors are paying for. It is also what makes the business defensible, because a competitor cannot reproduce years of collected footage overnight, and the labeling pipeline that turns raw video into training data is itself a substantial engineering achievement.
The revenue story
The most striking numbers in the Humyn Labs announcement are not the funding figures but the revenue figures. The company reported $10 million in revenue generated between April and October 2026, with an additional $20 million in signed deals and another $20 million under negotiation. It says it has 12 paying customers, including frontier AI labs, robotics companies, and developers of world models, with nine of those customers in the United States, one in Europe, and two in India. The company also claims to have surpassed $50 million in annualized revenue within six months of operations, a figure that extrapolates the recent run rate forward and should be read as the company's own claim rather than an audited result.
Those numbers deserve careful reading. Ten million dollars of revenue in seven months, from a company founded in February, is an exceptional start by any standard. Twenty million in signed deals suggests a pipeline that could triple the business if the contracts convert on schedule, and another $20 million under negotiation suggests demand beyond what is already signed. But the figures are company-reported and have not been independently audited in the available reporting, and the $50 million annualized figure is an extrapolation, not a realized number. Investors who led the round will have seen the underlying contracts and diligence, but outside observers should treat the revenue claims as directionally impressive and precisely unverified.
The customer concentration is also worth noting. Twelve paying customers is a small number, and nine of them are in the United States, which means the business depends on a handful of large buyers in a single market. That concentration is normal for an early-stage data company selling to frontier labs, and it is also a risk: losing one or two large customers would hurt. The counterweight is that those customers are among the best-funded companies in the world, they are spending aggressively on physical AI, and they have few alternative suppliers of the data Humyn Labs sells. For founders, the lesson is that a small number of large, committed customers can fund a business faster than a large number of small ones, provided the product is genuinely hard to replace.
Who is backing it and why
Accel's decision to lead the equity round is the strongest signal in the announcement. Accel is one of the most established venture firms operating in India, with a long history of backing companies that became category leaders, and its partner Pratik Agarwal is the named sponsor of the investment. When a firm of Accel's standing leads a $26 million equity round in a nine-month-old company, it is saying that it believes the team, the market, and the early traction are all real, and that the opportunity is large enough to justify the pace. For Humyn Labs, Accel's backing is also a practical asset: the firm's network across India, the United States, and Europe opens doors to customers, hires, and future investors.
The rest of the syndicate adds different kinds of value. Prosus Ventures brings the perspective of a global technology investor that has backed some of the largest consumer and marketplace companies in the world. Finserv Intelligence through Bajaj Finance and the Bajaj Alts AI Opportunities Fund bring the backing of one of India's largest financial groups, with deep relationships across Indian industry and a growing interest in artificial intelligence. Jump Trading, a quantitative trading firm, brings an investor that understands data businesses from the inside. Dallas Venture Capital, BeeGlobal, and Lighthouse Canton round out a group that spans venture capital, family offices, and debt.
Angel investor Anita Pramoda, the former chief financial officer of Epic, is a notable addition. Pramoda is described as deeply invested in the physical AI model stack, and her participation brings the perspective of someone who has managed the finances of one of the most successful technology companies in the world. For a company whose business model depends on the economics of data collection and whose customers are the largest AI labs, a chief financial officer of that caliber is a valuable sounding board. The syndicate as a whole reads as a deliberate assembly of investors who understand different pieces of the puzzle: the Indian market, the global AI ecosystem, the data business, and the financial discipline required to scale it.
The venture debt piece
The $21 million venture debt facility, led by Lighthouse Canton's Growth Debt Fund II with participation from Trifecta Capital, deserves its own attention, because it reveals how the business actually works. Humyn Labs pays its contributors when they record footage, and it gets paid by its customers on longer cycles, which creates a working capital gap: money goes out immediately and comes back later. A company that is growing fast faces a widening version of that gap, because each new customer and each new collection project requires more upfront payments before the revenue arrives. Venture debt is designed for exactly this problem, providing capital that bridges the timing gap without diluting shareholders.
The structure also tells investors something about the quality of the revenue. Lenders do not extend $21 million in venture debt to a nine-month-old company on the basis of a pitch deck. They do it on the basis of signed contracts, predictable cash flows, and a management team they trust to collect what it is owed. The debt facility is, in that sense, a second form of diligence: Lighthouse Canton examined the same business Accel did, and concluded that the cash flows were reliable enough to lend against. For founders, the lesson is that debt is not only for mature companies. When a business has contract-backed revenue and a timing gap between spending and collection, venture debt can be cheaper than equity and less dilutive, provided the revenue is genuinely predictable.
There is a risk, of course. Debt must be repaid whether or not the next round of deals closes, and a company that borrows against future revenue is betting that the future arrives on schedule. If the $20 million in signed deals slips, or if the $20 million under negotiation fails to convert, the debt still needs servicing. Humyn Labs' position is that the demand is strong enough and the pipeline deep enough to justify the leverage, and its lenders evidently agree.
What to watch next
The first thing to watch is whether the revenue claims hold up. Humyn Labs says it generated $10 million between April and October, has $20 million in signed deals, and has another $20 million under negotiation. If the signed deals convert and the negotiations close, the company will have demonstrated that its early traction was not a fluke but the start of a scalable business. If they slip, the $50 million annualized revenue claim will look optimistic, and the next fundraise will be harder. Watch for announcements of new customers, expansions of existing contracts, and any independent reporting on the company's financials.
The second is the deployment infrastructure. The funding is earmarked for building collection infrastructure in Brazil, South Korea, and India, strengthening the capture hardware and processing engine, and expanding the presence in San Francisco, London, and Seoul. Each of those is a measurable milestone: new hubs opening, new hardware shipping, new offices staffed. The third is the competitive landscape. Humyn Labs is not the only company collecting real-world data for physical AI, and the field includes well-funded startups, the data divisions of large technology companies, and simulation providers who argue that synthetic data can substitute for recorded footage. The question is whether real-world data proves decisively better, and whether Humyn Labs can collect it faster and cheaper than its rivals.
Angeled.com connects founders with investors.
Mint
BusinessFortnight
TradingView News
Inc42
YourStory
AF.net
StartupMars
Read Next
Why "Pre-Seed" Now Means $30M Checks in Robotics and Embodied AI
Arcus Biosciences Prices $250 Million Public Offering
Azoma Raises $11.2M Series A to Define Agentic Commerce Optimization
Dataline Raises $10.6M Seed to Build the Data Network for AI Agents
DeepSeek's $15 Billion Raise: China's Boldest AI Bet Yet
Electric Plant Company Raises $11.2M Seed to Read Plants' Electrical Signals With AI



