Mecka AI raises $60M to train robots on direct human data, targeting physical AI deployment

Mecka AI raises $60M to train robots on direct human data, targeting physical AI deployment

Mecka AI, a New York City-based startup founded in 2025, has raised $60 million in a two-part funding round — a $25 million Series A closed in November 2025 and $35 million in follow-on investments — led by Framework Ventures with participation from Menlo Ventures, Kindred Ventures, SV Angel, and Ted Xiao of Project Prometheus. CEO Josh Gao outlined the company's vision to become the "data and deployment layer for physical AI," using body sensors and iPhones to capture human motion data from homes, kitchens, and labs to train robots, departing from teleoperation and simulation-based approaches.

The funding and strategy place Mecka AI within the emerging robotics data infrastructure segment, where the scarcity of real-world training data is a key bottleneck. The company claims an annual run rate of $100 million from signed contracts, though it has not named customers. This revenue trajectory, achieved within roughly a year of founding, mirrors the fastest-ARR-ramp pattern seen in earlier AI infrastructure layers — suggesting that enterprise demand for physical AI training pipelines is materializing ahead of widespread robot deployment.

Mecka AI's data-first approach to robotics training fits the acqui-licensing and data-moat dynamics that have defined earlier AI substrate layers: controlling the proprietary human dataset becomes the defensible asset, rather than the robot hardware or simulation environment itself. With 40 employees and plans to expand its data engineering and research teams, the company is betting that direct human data will prove more reliable for commercial robot tasks than synthetic or simulated alternatives — a thesis that, if validated, could reshape how the physical AI stack is built.