Indian physical AI startup Humyn Labs has raised $47 million in Series A funding led by Accel to expand its real-world data collection infrastructure, robotics research capabilities and deployment operations. The financing comprises $26 million in equity and $21 million in venture debt, as the Bengaluru-based company looks to help AI laboratories and robotics developers train machines to perform tasks in real-world environments.
Humyn Labs is building a business around collecting human activity data from factories, warehouses, farms and homes, then converting that information into structured datasets used to train AI models. Founded by Manish Agarwal and Ishank Gupta, the startup plans to expand infrastructure in India, Brazil and South Korea while building capabilities to test robots in real-world environments, marking a shift beyond traditional AI data-labeling services towards physical AI research and deployment.
Key takeaways
- Humyn Labs raised $47 million in Series A funding.
- Accel led the $26 million equity round.
- Lighthouse Canton led the $21 million venture debt component, with Trifecta Capital participating.
- Investors include Prosus Ventures, BeeGlobal, Bajaj Finserv-linked entities, Dallas Venture Capital and Jump Trading, among others.
- The startup captures human activity across more than 20 countries to build training data for physical AI and robotics.
- Humyn Labs said annualised revenue crossed $50 million within six months of operations, according to Moneycontrol.
- TechCrunch separately reported $10 million in revenue from April to October, $20 million in signed contracts and another $20 million in deals still under negotiation.
- The startup plans to expand robotics deployment infrastructure in Brazil, South Korea and India.
Humyn Labs raises $47 million in Series A funding
The $47 million financing is split between equity and debt, giving Humyn Labs capital to expand its operations while managing the working-capital requirements of its data collection business.
The equity portion totals $26 million and is led by Accel. Participants include Prosus Ventures, BeeGlobal, Finserv Intelligence, Bajaj Alts AI Opportunities Fund, Dallas Venture Capital, Lighthouse Canton, Jump Trading and former Epic CFO Anita Pramoda, according to Moneycontrol and Business Standard.
The remaining $21 million comes through venture debt led by Lighthouse Canton, with participation from Trifecta Capital.
Venture debt is financing provided to startups in the form of loans rather than equity investments. It allows companies to raise capital without immediately diluting the ownership of existing shareholders, although the debt must eventually be repaid and may carry interest, fees and other obligations.
Humyn Labs operates a business that requires significant upfront spending. It pays data collectors and field partners for gathering information, processing and validating recordings, and preparing datasets for AI customers.
Its customers, meanwhile, can take 30 to 60 days to pay invoices, according to co-founder Manish Agarwal’s comments reported by TechCrunch.
The venture debt is intended partly to bridge this payment gap, allowing the company to fund operations while waiting for customer payments.
What Humyn Labs actually does
Humyn Labs focuses on a data problem facing robotics and physical AI companies: machines need to learn how humans interact with physical objects and environments.
AI models trained primarily on text, images or digital information may understand descriptions of objects and tasks but still struggle to perform physical actions reliably.
A robot operating in a factory, warehouse, kitchen or farm must understand much more than the names of objects. It needs to recognise where things are, how they move, how people manipulate them, how different actions affect the environment and what to do when conditions change.
Humyn Labs captures first-person human activity, including how people see, navigate and interact with objects while performing everyday work.
The company uses specialised data-capture hardware to record activities and then processes the footage through software that extracts information such as hand movements, object interactions and task sequences.
This information can be converted into structured datasets that robotics and AI researchers use to train or evaluate their systems.
<text weight=”medium”>The process is broadly divided into four stages:</text>
- Capture: Record real-world human activity using cameras and other sensors.
- Processing: Clean, organise and synchronise the collected information.
- Annotation: Label actions, objects, movements and relevant details.
- Training and evaluation: Deliver datasets to AI labs and use real-world testing to identify where robotic systems need improvement.
The objective is to provide AI developers with data that reflects actual physical environments rather than only simulations or curated demonstrations.
Why robots need human data to learn physical tasks
Physical AI refers to artificial intelligence systems that perceive, reason about and act in the physical world. It includes industrial robots, autonomous machines, humanoid robots, robotic assistants and other systems designed to perform physical tasks.
The development of these machines requires a different kind of training information from conventional chatbots.
A language model can learn from books, websites, articles and other text-based information. A robot, however, must understand movement, spatial relationships, contact with objects, force, timing and environmental changes.
For example, folding a piece of clothing is a relatively simple task for a human but a complicated problem for a robot. The fabric changes shape, its position can shift, and the robot must adjust its grip and movements continuously.
Similarly, picking up a fragile object requires understanding how much force to apply. Opening a door, assembling a component or sorting items in a warehouse requires physical interaction and feedback.
Robotics companies need large quantities of diverse examples to train models that can handle such situations.
Humyn Labs is attempting to address this gap by capturing how humans perform tasks across different environments and converting the activity into usable AI training data.
From data collection to real-world robot deployment
Humyn Labs is positioning itself beyond the traditional role of a data supplier.
Instead of only collecting footage and delivering datasets, the company wants to help AI and robotics developers test systems in real-world settings and compare machine performance with human activity.
This creates a feedback loop.
Human demonstrations provide information about how tasks are performed. AI models learn from the data, robots attempt the tasks, and testing helps reveal where the systems fail or need additional training.
Those findings can then inform new data collection and model development.
The company says it works with three broad categories of customers: frontier AI laboratories, robotics labs developing vision-language-action models, and companies building world models.
Vision-language-action models, or VLA models, combine visual information, language instructions and action generation. They are designed to help robots interpret what they see, understand what they are being asked to do and determine appropriate actions.
World models aim to represent how environments behave and how they change over time, allowing AI systems to reason about possible actions and their consequences.
Humyn Labs wants its data and deployment infrastructure to support both types of development.
Where Humyn Labs collects data
The startup operates data-capture and enrichment programmes across more than 20 countries, spanning Latin America, Africa, East Asia, South Asia and Southeast Asia, according to Moneycontrol.
Its data collection includes environments such as:
- Factories and manufacturing facilities
- Warehouses and logistics operations
- Farms and agricultural settings
- Residential homes
- Commercial workplaces
- Everyday human activities involving objects and movement
The diversity of these environments matters because robots must eventually operate outside tightly controlled laboratory conditions.
A robot trained only in one factory layout may struggle when it encounters different equipment, lighting, object arrangements or movement patterns.
Likewise, data collected in a single country or environment may not represent the variety of physical settings robots encounter globally.
Humyn Labs is betting that collecting information from different environments, people and tasks can improve the diversity of training datasets available to AI developers.
The startup has identified Brazil, South Korea and India as important markets for expanding its deployment-readiness infrastructure.
How the $47 million will be used
Humyn Labs plans to use the new capital to expand its data collection capabilities, build infrastructure for real-world robotics deployment and strengthen its research and customer-facing operations.
Its investment plans include:
| Investment area | Planned objective |
|---|---|
| Robotics deployment centres | Prepare and test robots in real-world environments |
| India, Brazil and South Korea | Expand deployment-readiness infrastructure |
| Data capture hardware | Improve how human activity and physical interactions are recorded |
| Data enrichment platform | Process, validate and label collected information |
| Research capabilities | Develop new data formats and capabilities across audio, video, motion and touch |
| Global hubs | Strengthen research and customer-facing teams in international markets |
The company is also planning to expand its audio and video datasets while developing new capabilities involving motion and touch.
The expansion could help Humyn Labs move closer to becoming an infrastructure provider for physical AI developers rather than a company focused only on collecting raw data.
Humyn Labs claims more than $50 million in annualised revenue
The funding announcement is accompanied by a rapid growth claim.
Co-founder Manish Agarwal told Moneycontrol that Humyn Labs crossed $50 million in annualised revenue within six months of operations.
Annualised revenue is a run-rate estimate. It takes revenue generated over a particular period and extrapolates it to a full year, assuming the pace continues.
It is not the same as revenue actually earned during a full financial year.
TechCrunch reported a different set of revenue figures from the company: $10 million in revenue between April and October, $20 million in signed contracts and another $20 million in deals that had not yet been finalised.
These numbers are not necessarily contradictory because they measure different things.
| Metric | Reported figure | What it means |
|---|---|---|
| Funding raised | $47 million | Equity plus venture debt |
| Equity funding | $26 million | Capital invested for ownership |
| Venture debt | $21 million | Borrowed capital |
| Revenue, April–October | $10 million | Reported realised revenue during that period |
| Signed contracts | $20 million | Contracted business reported by the company |
| Deals under negotiation | $20 million | Potential business not yet finalised |
| Annualised revenue | More than $50 million | Reported revenue run rate, not full-year realised revenue |
Investors will need to distinguish between actual revenue, contracted revenue, pipeline and annualised revenue when assessing the startup’s financial performance.
The company’s rapid growth claim, if sustained, could indicate strong demand for real-world data and robotics-related services. But its future performance will depend on customer retention, delivery quality, contract execution, operating costs and the ability to scale its data collection infrastructure.
Why venture debt is important for this business
The $21 million venture debt component is particularly relevant because Humyn Labs has a cash-intensive operating model.
The company needs to pay data collectors, local partners, equipment providers and other operational expenses before it necessarily receives money from its customers.
That creates a working-capital gap.
Working capital is the money a business needs to cover day-to-day expenses while waiting for cash to come in from customers.
In a data collection business, the company may have to pay workers or contractors weekly while customers settle invoices several weeks later.
Venture debt can help fund this gap without requiring the startup to raise additional equity immediately.
However, debt also introduces financial obligations. Humyn Labs will need to generate sufficient cash flow to repay its borrowing and manage the cost of financing.
The balance between equity funding, debt and operating cash flow could therefore become important as the company expands internationally.
India’s opportunity in physical AI
Humyn Labs’ expansion could also place India in a growing position within the physical AI ecosystem.
India has a large workforce across manufacturing, agriculture, logistics, construction, retail and other industries where physical tasks are performed daily.
Those environments can provide opportunities to collect data representing how people perform tasks, manipulate objects, navigate spaces and interact with equipment.
India could become relevant not only as a market for robotics deployment but also as a source of diverse real-world training data.
The country’s linguistic diversity, varied work environments, industrial facilities and everyday household activities could be useful for training AI systems designed for global markets.
However, collecting data in India does not automatically make it valuable. Data quality, consistency, consent, annotation accuracy, representation and the ability to convert information into usable training formats are crucial.
Humyn Labs’ business will depend on whether it can turn large-scale collection into datasets that AI developers actually find useful.
The startup’s planned expansion in India could create opportunities in field operations, data engineering, AI research, hardware development, robotics testing and dataset quality control.
Its success will also depend on building systems that can manage data collection and processing at scale without allowing costs to grow faster than revenue.
The privacy and consent challenge
Collecting first-person video and human activity data across homes, factories and workplaces raises privacy considerations.
Videos recorded in residential or commercial environments can contain sensitive information, including faces, personal belongings, workplace details, confidential processes and other identifiable information.
Humyn Labs has said it works with local partners to identify collection sites and supervise operations. Co-founder Manish Agarwal told TechCrunch that participants record their regular work or household activities rather than staged tasks.
The company’s business therefore requires clear processes for consent, privacy protection, data handling, access control and the removal or management of sensitive information.
As robotics training data becomes more valuable, the quality of governance surrounding data collection could become an important differentiator.
AI developers increasingly need not only large datasets but also information about how the data was collected, whether participants consented, how privacy risks were handled and whether the data is suitable for commercial or research use.
The competitive landscape for physical AI data
Humyn Labs is entering a market that includes companies collecting human activity, developing robotics datasets, building simulation environments and testing robotic systems.
The market differs from traditional AI data services because physical AI requires more than text annotation or image labelling.
Robotics developers may need first-person video, hand movement tracking, object interaction data, sensor streams, audio, spatial information and task-specific demonstrations.
Companies such as Human Archive, Neocambrian AI and FPV Labs are also operating in areas related to real-world human data and physical AI, according to Moneycontrol’s reporting.
The competitive advantage may ultimately depend on the diversity of data, the quality of the collection process, hardware capabilities, geographic coverage and the ability to translate raw data into useful training material.
Humyn Labs believes its opportunity is larger than simply gathering data. It wants to build the infrastructure that connects real-world human activity with AI model development and robot deployment.
The bigger picture
The next phase of AI development may increasingly involve systems that can operate outside computers and digital interfaces.
AI is already generating text, images, videos, code and software applications. The next challenge is teaching intelligent systems to navigate physical environments, understand objects and perform useful actions safely and reliably.
That requires a different infrastructure layer.
Humyn Labs is betting that human experience can become a foundational input for this development. Its business combines data capture, processing, enrichment, research and real-world testing.
The $47 million funding round gives the startup capital to expand this infrastructure while increasing its international reach.
The opportunity is significant, but execution will be difficult. Collecting diverse data across countries, maintaining quality, building hardware, managing privacy and developing robotics deployment centres require substantial operational capabilities.
The company will need to demonstrate that its data and research infrastructure can improve robotic performance and become a repeatable, profitable business.
What happens next?
Humyn Labs plans to expand its real-world deployment infrastructure across Brazil, South Korea and India while strengthening data capture, processing and research capabilities.
The company also intends to develop additional capabilities involving motion and touch, extending its data collection beyond existing audio and video operations.
Its next phase will likely focus on converting its global data collection network into a more integrated physical AI research and deployment platform.
For investors, the key metrics to watch will be revenue conversion, signed contracts, customer concentration, data quality, operating margins, working-capital requirements and the pace of robotics deployment.
For India’s AI ecosystem, Humyn Labs represents a different category of startup opportunity: building the infrastructure, data and physical-world capabilities that future AI and robotics systems may depend on.
The company is not manufacturing humanoid robots. It is building the human-data and deployment layer that could help those robots learn how to work in the real world.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.



