India's Workforce Fuels Robot Training Data Boom for AI Companies

India emerges as robot training hub

India may lag behind the U.S. and China in manufacturing AI-enabled robots, but the country has carved out a distinctive role in the physical AI supply chain: providing the human labor needed to train robots through demonstration data.

Workers across India are now recording first-person videos of routine activities—cooking, cleaning, packing—to generate training datasets for robotics companies. These recordings teach robots how to perform tasks in real-world environments, a critical step in developing machines that can operate outside controlled laboratory settings.

Tanisha Reddy, a teacher in southern India, earns less than $4 per hour recording 3-4 hours of daily video for Qanat Consulting Services. The work fits around her primary job and childcare responsibilities, requiring no specialized skills beyond performing ordinary household tasks while wearing a head-mounted camera.

Why it matters

The humanoid robot market is projected to reach $200 billion within a decade, according to Barclays, with Morgan Stanley forecasting it could surpass $5 trillion by 2050. India's position as a data supplier mirrors its earlier role in IT services, but experts warn the country must move beyond commoditized data collection to capture lasting value. As contract prices have already halved due to competition, Indian firms face pressure to develop proprietary datasets and conversion capabilities rather than simply executing collection tasks for foreign clients.

From collection to conversion

Multiple data collection firms have launched in India within the past year, recruiting workers to record egocentric videos on behalf of U.S. and Chinese robotics companies. Thaslim Pattan, founder of Qanat Consulting Services, reports winning contracts from garment manufacturers and other industrial clients, though she notes intensifying price competition as new entrants flood the market.

Some Indian startups are attempting to move up the value chain. Neocambrian AI opened a robotics data factory in Noida last month and maintains a network of over 100 factories where workers record task performance. Rather than collecting data on demand, the company builds proprietary datasets focused on teaching robots object manipulation and dexterity—capabilities that require an estimated 100 million hours of video to approach human-level performance.

Abhinav Kukreja, Neocambrian AI's founder, believes India can dominate this layer of the AI stack by becoming the "human labor marketplace of the world." His firm retains ownership of the datasets it creates, positioning itself as a data provider rather than a contract collector.

Humyn Labs takes a similar approach, collecting data from India (35%), Latin America (50%), and elsewhere in Asia (15%) while focusing on data conversion and diversified environments. Co-founder Manish Agarwal argues that India must "evolve from collector to converter" as the data collection market inevitably saturates.

The commoditization challenge

Industry observers acknowledge that basic data collection work is rapidly becoming commoditized. While India's large workforce and low labor costs provide a current advantage, maintaining relevance will require developing expertise in data processing, quality verification, and dataset ownership—capabilities that command higher margins than simple video recording.

The parallel to India's IT services industry is instructive: the country succeeded by moving beyond basic outsourcing to offer higher-value engineering and consulting services. Whether Indian robotics data firms can replicate that trajectory remains an open question as global demand for training data accelerates.

These details were first reported by CNBC in its "Inside India" newsletter.