Mecka AI, a startup that collects and analyzes human motion data to train humanoid robots and other machines, is nearing a new funding round led by Sequoia Capital at a valuation of roughly $500 million, according to a TechCrunch report citing two people with knowledge of the deal. The financing would come just three months after the company announced a $60 million round led by Framework Ventures.
The exact size of the new round has not been disclosed, and TechCrunch noted that terms are not final and could still change. Mecka AI did not respond to a request for comment, and Sequoia declined to comment. For more on the money flowing into AI startups, follow our startup and venture capital coverage.
From Fintech Founders to Robot Data
Mecka AI was co-founded in 2024 by four entrepreneurs: Canadians Josh Gao and Mogen Cheng, who previously built a restaurant fintech startup; Jason Chong, who joined Coinbase after it acquired his crypto exchange; and Duy Nguyen, who focuses on operations. None of the four come from robotics — but they spotted a bottleneck that has become one of the industry's most valuable problems.
The company's name derives from "mecha," the fictional genre of giant robots controlled by humans, and its business model follows the same logic that built the data-labeling giants of the language model era. Mecka pays people to record themselves performing everyday tasks — making coffee, fixing cars — using body sensors and smartphones, capturing what the industry calls "egocentric" data: real-world human motion from the performer's point of view.
That data is then sold to robotics companies and AI labs training general-purpose and humanoid robots, alongside other physical data collection methods such as teleoperation. In effect, Mecka is trying to do for robotics what Scale AI, Mercor, Surge and other human data companies did for large language models.
Revenue Growth Caught Sequoia's Attention
The financial trajectory explains the premium valuation. As of early June, Mecka was projecting it would end 2026 at an annual run rate of $100 million, co-founder Josh Gao told Fortune when the company announced its previous fundraise. For a company founded in 2024, reaching a nine-figure revenue run rate within two years — if achieved — would put it among the fastest-scaling companies in the physical AI supply chain.
The $60 million round announced in June was led by Framework Ventures with participation from Menlo Ventures, SV Angel and Kindred Ventures. A Sequoia-led round at roughly eight times that valuation inside three months would mark one of the steepest valuation climbs in the sector this year.
The Physical Data Gold Rush
Mecka is not alone in betting that real-world data is the scarce resource of the robotics era. TechCrunch reported last week that XDOF, another startup collecting real-world data for robot training, is nearing a new round at a $1.2 billion valuation. Meanwhile, human-data platforms built for LLMs — including Scale AI and Micro1 — are expanding beyond language data into physical-world collections.
The rush reflects a structural gap: while text and images exist on the internet in nearly unlimited quantities, nobody has recorded the physical world at scale. Humanoid robot makers need millions of hours of demonstrations of humans doing ordinary tasks — folding laundry, assembling furniture, operating machinery — to teach their machines dexterity, balance and judgment. Capturing that data with motion sensors and cameras is currently one of the only scalable methods, which is why investors are pricing these companies like infrastructure rather than tools.The egocentric approach Mecka uses has distinct advantages over alternatives. Teleoperation, where a human pilots a robot arm remotely and logs the movements, produces data tied to a specific robot's geometry, which makes it expensive to reuse across hardware generations. Motion capture from wearable sensors, by contrast, is hardware-agnostic: a recording of a mechanic rebuilding an engine is useful whether the customer trains a humanoid, an industrial arm, or a quadruped. It is also far cheaper to scale, since anyone with a smartphone and a sensor kit can become a contributor — the same crowdsourcing logic that let labeling platforms assemble workforces of thousands overnight.
The economics also mirror the language-model era in a less comfortable way. Data companies in the LLM boom discovered that their customers — the labs — are also their most capable potential competitors, with the resources to build collection arms in-house. Robotics labs are already experimenting with simulation and video-learning pipelines to reduce their dependence on purchased demonstrations. That tension between recurring data demand and customer disintermediation will shape whether today's datavaluations hold.
What to Watch
The deal has not closed, and both parties declined to comment on the record. But if it completes near the reported terms, the robotics data sector will have produced two nine-figure valuations in a single week, confirming that the demand shock has moved from humanoid hardware — where companies like Figure, Tesla and Unitree compete — one layer down to the data that trains them.
It also raises the question every data intermediary eventually faces: whether the labs buying the data will keep paying intermediaries, or bring collection in-house as the market matures. For now, the money says the bottleneck is real and worsening — and that investors are willing to pay up for whoever closes it.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
