Mecka AI, a startup whose entire business model is paying humans to demonstrate their own replaceability, is approaching a $500 million valuation in a new round led by Sequoia Capital. The previous round, a modest $60 million, closed just three months ago. The machines are in a hurry.
The company pays people to record themselves making coffee and fixing cars — footage that will one day be used to determine whether a person is still needed for either task.
What happened
Mecka AI collects human motion data — people performing everyday tasks using body sensors and smartphones — and sells it to robotics companies and AI labs training humanoid robots. The startup named itself after "mecha," a genre of fiction featuring giant robots controlled by humans. The name, in retrospect, is doing a lot of optimistic work.
The company was founded in 2024 by four entrepreneurs whose backgrounds include restaurant fintech and a crypto exchange. None of them have a background in robotics. They did, however, notice that robots couldn't learn to navigate the physical world without watching humans do it first, which is the kind of insight that turns out to be worth half a billion dollars.
As of June, Mecka was projecting $100 million in annualized revenue by year-end. At that rate, the humans filming themselves are, by any reasonable measure, very good at their jobs. For now.
Why the humans care
The scramble for physical-world training data has become the central bottleneck in humanoid robotics — a field in which every major lab is now competing to build a general-purpose robot capable of operating in unstructured human environments. Mecka's competitors include XDOF, currently approaching a $1.2 billion valuation, and Scale AI, which is expanding from language models into the physical world with what can only be described as commitment.
The company pays contributors to perform tasks like making coffee, fixing cars, and presumably other things humans still do without assistance. This data, captured from a first-person "egocentric" perspective, trains the models that will eventually not need the footage anymore. The arrangement is efficient. It is also the most literal possible definition of a knowledge transfer.
What happens next
The terms of the Sequoia deal are not yet final. More rounds will follow, because the demand for human demonstration data will continue until the demonstrations are no longer necessary.
The humans filing into Mecka's platform to record themselves performing everyday tasks are, in the most technical sense, doing exactly what they were hired to do.