Most AI startups raise money to build a model. Mecka raised a $60M Series B to build the thing the models are starving for: a record of how humans actually move, grip and apply force in the physical world. CEO Josh Gao's framing of the gap is the whole thesis in one line — "Motion, contact, force and geometry aren't on the internet."

The round is led by Sequoia Capital, with a roster of strategic backers that reads like a who's who of the chips, cloud and robotics stacks. And the number that should make you look twice isn't the $60M. It's the revenue figure sitting underneath it.

$60M
Series B, led by Sequoia
$120M+
Total raised, per the company
$100M
Run-rate revenue, mid-year (self-reported)
$300M
Run rate Mecka projects by year end

The Data Layer, Not the Robot

Everyone in physical AI is chasing the same scarce resource. Large language models were trained on text that already existed on the open internet; there is no equivalent corpus for how a hand closes around a mug or how weight shifts when you lift a box. The economic scenarios AI labs are already modelling assume AI keeps reaching into physical work, and that reach runs straight into this data gap. Gao's company says it went and made one. In his words, the raise "brings our total funding to over $120M to become the data and deployment layer for physical AI."

That positioning matters. Mecka is not pitching itself as another humanoid-robot company competing on hardware. It is pitching itself as the supplier one layer below — the place robot builders go for the motion, contact, force and geometry data their models can't learn from scraped video. In a gold rush, this is the sell-shovels play.

How does it get the data? Not by downloading it. Mecka says it designs and manufactures its own multi-sensor hardware for each signal class, then runs what it calls a global production operation, with capture fleets recording human demonstration in homes and commercial environments. The bet, in the company's telling: "We bet that real-world human demonstration was the key to scaling." So "we went out and recorded the world."

A Revenue Curve That Doesn't Look Like a Series B

Here is the line that reframes the whole round. Mecka says it surpassed $100 million in run-rate revenue in June — a self-reported figure, and one it says it hit within only a few months of operations. From there, the company projects a $300M run rate by the end of the year.

Treat the $300M as a projection, because that is what it is — a forward target the company has set itself, not money in the door. But even the backward-looking $100M number is unusual for a company still labelling its raise a Series B. Most startups at the B stage are proving that demand exists. Mecka is claiming the demand is already a nine-figure annual business and accelerating.

Both revenue figures come from Mecka itself and have not been independently audited. Run-rate revenue annualizes a recent short period, so it reflects a current pace rather than realized annual sales — and the $300M is an explicit projection, not a result. Read both as the company's own claims.

If the self-reported numbers hold up, they explain the investor list better than any pitch deck could. Strategic money doesn't usually chase a pre-revenue science project at this scale. It chases a supplier that the whole ecosystem may end up depending on.

Who Wrote the Checks

Sequoia leads, but the supporting cast is the tell. According to Mecka, the round is backed by new investors including Nvidia, M12 (Microsoft's venture fund), Qualcomm Ventures and Samsung, with continued support from Kindred, Framework Ventures and Neo.

Look at what those four new names have in common. Nvidia sells the compute robotics training runs on. Qualcomm and Samsung sit on the chip and device side of any edge-robotics stack. Microsoft, whose CEO has argued you end up paying for AI twice, is the cloud and enterprise distribution layer. These are not generic growth funds parking capital — they are the companies that would each benefit if a dominant data layer for physical AI emerged, and each now has a stake in the one claiming the position.

The angel list reinforces the same operator-heavy signal. Mecka names three: "Tony Xu, founder and CEO of DoorDash; Frank Slootman, former CEO of ServiceNow and Snowflake; Milan Kovac, former head of Optimus at Tesla." Xu built a physical-world logistics machine, Slootman is the archetypal enterprise-scaling CEO, and Kovac ran a humanoid program at Tesla. That is operational pattern-matching across logistics, enterprise go-to-market and robotics itself.

Role in the roundWho
LeadSequoia Capital
New strategic investorsNvidia, M12 (Microsoft), Qualcomm Ventures, Samsung
Continuing investorsKindred, Framework Ventures, Neo
AngelsTony Xu (DoorDash), Frank Slootman (ServiceNow/Snowflake), Milan Kovac (ex-Optimus, Tesla)

The Academic Validation Lane

Capital proves that investors believe. It doesn't prove the data is any good. For that, Mecka points to research use: it says it tested its bet with researchers at Georgia Tech, Stanford, UC San Diego, ETH Zurich, MIT and Meta through a dataset it calls EgoVerse.

That is a sensible proof mechanism. If independent labs at that tier are building on your data, it is harder to dismiss the capture approach as a demo. It is worth noting the direction of the claim, though: academic use shows the data is useful to researchers, which is related to, but not the same as, proving it moves the needle on commercial robot performance. The $100M run-rate claim is the stronger evidence on that front, if it holds.

The Question Nobody in the Announcement Answers

A business whose core activity is recording human demonstration in people's homes and commercial spaces raises obvious questions about consent, compensation and how the footage is governed. The announcement describes the capture fleets and the hardware; it does not spell out the arrangements with the people being recorded.

That's not an accusation — there's no evidence in the public record of mishandling, and none should be inferred. But it is the open question that follows a company scaling physical-world data collection this fast. As the dataset grows into a nine-figure business, how contributors are recruited, paid and protected is the part of the story most likely to draw scrutiny next. Investors bet on the revenue curve; regulators and the public will eventually ask about the supply chain behind it.

What to Take Away

  • The product is the dataset, not a robot. Mecka is selling the data and deployment layer for physical AI — the motion, contact, force and geometry that, in Gao's words, "aren't on the internet." Sell shovels, don't dig.
  • The revenue claim is the real headline. A self-reported $100M run rate reached mid-year, with a projected $300M by year end, is what separates this from a typical Series B — treat the projection as a target, not a result.
  • The investor list is strategic, not generic. Nvidia, Microsoft's M12, Qualcomm and Samsung each sit on a different part of the robotics stack and each now has a stake in the data layer feeding it.
  • Operator angels signal the playbook. DoorDash's Tony Xu, Snowflake's Frank Slootman and Tesla's former Optimus lead Milan Kovac point at logistics, enterprise scaling and humanoid robotics all at once.
  • Watch the supply chain, not just the balance sheet. How a human-motion data company handles consent and compensation for the people it records is the unanswered question most likely to define its next chapter.

Analysis and opinions are BougainWell's own. Revenue and funding figures are as reported by Mecka and its CEO and have not been independently verified.

Sources

Cover image: Mecka, via its Series B announcement at mecka.ai/news/series-b.

Reporting based on Mecka's own Series B announcement and its CEO's post. All framing and conclusions are BougainWell's own.