Sequoia-led round values Mecka AI near $500M for footage of people making coffee

Somebody is standing in a kitchen with sensors on their arms and a phone propped against the kettle, making coffee slowly enough for a camera to follow. Mecka AI pays them to do it, and pays other people to record themselves fixing cars ^[8]. Sequoia Capital is leading a round that prices the company doing the paying at about $500 million, according to two people who spoke to TechCrunch ^[1].

Josh Gao told Fortune in early June that Mecka projected ending 2026 at a $100 million annual run rate ^[9], which puts about $500 million at roughly five times that projection ^[13]. The June financing was $60 million led by Framework Ventures, with Menlo Ventures, SV Angel and Kindred Ventures ^[2]. That is a round size, not a valuation, so the two marks do not yield a step-up. TechCrunch reported that it has not learned the size of the new round and that the terms could still change ^[3].

The pitch is Scale AI for robots, and the company says so itself: it set out to do for robotics what Scale AI, Mercor, Surge and other human data companies did for language models ^[7]. The work underneath is recruitment, sensor logistics and payouts. The founding team fits that description. Gao and Mogen Cheng previously built a restaurant fintech startup, Jason Chong joined Coinbase after it acquired his crypto exchange, and Duy Nguyen handles operations ^[5]. None of the four came out of robotics; what they identified, per TechCrunch, was a dearth of physical-world data and a conclusion that capturing real-world interactions was the primary bottleneck for general-purpose robots ^[6].

That conclusion is the founders' own. No results have been published to support it. What is public is investor behaviour. TechCrunch reported last week that XDOF, another real-world data collector, was nearing a round at a $1.2 billion valuation ^[11], about 2.4 times Mecka's reported mark ^[14], and that human-data platforms including Scale AI and Micro1 are moving into the same territory ^[12]. Mecka has not publicly disclosed its customer list, and robotics companies and AI labs that use this egocentric capture pair it with other methods such as teleoperation ^[10]. The pairing means human-worn capture supplements data recorded on the target hardware.

So a team weighing an hours purchase has two tests, neither of which the round settles. The first is whether a fixed block of hours moves success rate on a task the team has already benchmarked on its own robot. The second is what re-collection costs when the gripper, the workcell or the task changes, because if the answer is the same price again, those hours are a consumable and the vendor's run rate is your recurring line item. Mecka AI did not respond to a request for comment, and Sequoia declined to comment ^[4].