Humyn Labs expands data collection for physical AI with $20 mn push

AI firm Humyn Labs said in a press release on Monday that it has committed $20 million to expand its data collection operations in India, Southeast Asia, Latin America and West Asia.

The capital will finance the infrastructure needed to train physical artificial intelligence systems (robots) and voice models. Co-founders Manish Agarwal and Ishank Gupta are channeling the investment into organizing and validating human intelligence, the statement said.

Agarwal said. Mint Still in its early stages and not backed by any corporate or venture capital, Humyn is using its proceeds to invest $20 million in building high-quality datasets for physical AI, with a focus on egocentric and speech-based voice data. In AI, egocentric refers to data obtained from a first-person perspective when a human or agent interacts with its environment.

The firm focuses on source-first data collection, recording first-person human activities, visuals and movements. This data collection occurs in commercial, agricultural and residential environments. Agarwal explained that datasets capture how people navigate and physically interact in the environment to train physical AI systems.

The company is expanding its voice data infrastructure to include 33 languages, dialects, accents and code-switching patterns. This expansion addresses the use of voice for real-world commands and human-robot interactions.

Humyn Labs will establish robotics laboratories to create simulation environments and world models. This unit combines real-world data with training frameworks to deploy physical AI systems. In AI systems, the world model is the internal representation of how the real world works. For robotics labs, these world models are often used to allow robots to learn in simulation before actual deployment.

Humyn AI uses a decentralized network across the global south to deliver data. The current pipeline is driven by approximately 15-18 customers. While Agarwal did not disclose any client names, he said he is targeting top-tier labs where successful proofs of concept can be quickly scaled to $10-15 million contracts.

The company’s growth is currently uneven. “Though we have generated around $2 million in revenue in the last few months, we are operating at a run rate of around $4-5 million annually,” Agarwal said.

Agarwal said Humyn has a sales pipeline of around $45-50 million. “Based on current execution, we expect to achieve Annual Recurring Revenue (ARR) of $100 million by the end of December 2026.”

According to market research firm Fortune Business Insights, the global AI training dataset market size is estimated to be worth $3.59 billion by 2025 and will grow from $4.44 billion in 2026 to $23.18 billion by 2034, while the Indian market is estimated to be worth $190 million by 2026.