Mecka AI has closed a $60 million Series B led by Sequoia Capital for robot training data, Mecka says in its Series B announcement. Mecka AI is a data supplier that pays people to record physical work used to train humanoid and other robots. Mecka says it already supplies several top frontier robotics labs and multiple Mag 7 companies. It says it passed $100M in run-rate revenue in June 2026 and projects $300M by year-end.
Table of Contents
What Mecka records
Mecka pays people to record everyday tasks while wearing body sensors and using smartphones, TechCrunch reports in its startup funding report. The method captures ordinary work in its normal setting, not in a lab.
Examples named by TechCrunch include: For automation teams, that means training clips show clutter, timing and hand contact. Such variation matters when a robot must repeat a task outside a fixed cell.
- laundry folding and handling
- coffee-making
- car repair
Who joined the round
New backers include NVIDIA, M12–Microsoft's Venture Fund, Qualcomm Ventures and Samsung, Mecka says. Existing investors Kindred, Framework Ventures and Neo also joined.
Angels include DoorDash CEO Tony Xu, ex-ServiceNow/Snowflake CEO Frank Slootman and ex-Tesla Optimus head Milan Kovac, Mecka says. TechCrunch had previously reported the round was nearing close at about a $500 million valuation.
Why motion data stays scarce
Motion, contact, force and geometry are not on the internet and cannot be scraped or licensed, Mecka says. Robots therefore still need costly real-world demonstration capture.
That constraint shapes buying decisions. Teams cannot wait for web-scale data to solve manipulation. They must budget for collection, cleaning and task-specific coverage.
Where the funding goes
The capital will fund more data-capture hardware, expanded research and increased commercial robot deployments that post-train on site data, AI Insider reports in its funding coverage. Post-training means the base behavior is refined with data from the deployment site.
For operators, the practical point is deployment-specific data. A picking, wiping or assembly skill still needs local examples before it runs reliably.
