Prince Mario-Max Schaumburg-Lippe: Why Robots Are Learning From Coffee Drinkers, Not Code

Somewhere right now, a person is pouring a cup of coffee while wearing a sensor suit, and that small, boring moment might be worth more than a warehouse full of robot hardware. The Toronto startup Mecka just proved it. The two-year-old company closed a $60 million Series B led by Sequoia Capital, and TechCrunch reported the round prices Mecka at $500 million. The investors joining Sequoia are the interesting part: Nvidia, Microsoft’s M12 fund, Qualcomm Ventures, and Samsung all took part. Four hardware giants bought into a data company.

That tells you where the money thinks the bottleneck is. Robots are getting cheaper to build. Teaching them to move is the expensive part. As the company puts it: motion, contact, force, and geometry are not on the internet. Somebody has to perform them first.

The founders took the long way around

None of Mecka’s four founders came from robotics. CEO Josh Gao and Mogen Cheng sold a restaurant-payments startup in 2023. Jason Chong ran a crypto exchange that Coinbase acquired. Duy Nguyen built a business reselling sneakers. By their own admission, nobody at a robotics lab would have hired this crew on credentials alone.

So they did something old-fashioned. They spent months reading robotics papers and touring labs, asking everyone the same question: what is actually slowing this industry down? The answer kept coming back the same. Text and images could be scraped by the terabyte. Physical skill could not. The robot industry was starving for the one thing no cloud could offer: records of how hands really work.

1,362 hours of watching people be people

Mecka’s answer is the EgoVerse dataset, and the numbers are already serious. It holds 1,362 hours of recorded demonstrations across 80,000 episodes, performed by 2,087 unique demonstrators. People wear body sensors and use smartphones to film everyday tasks from their own point of view: making coffee, fixing a car, folding laundry, preparing a beverage. The sensors capture joint angles, force, and timing at 200 frames per second, and the company’s reconstruction models track hands with sub-centimeter accuracy.

This is a deliberate break from the industry’s usual method. Most robot training data comes from teleoperation, a person remotely steering a robot while it records. Mecka flips it. A human reaching for a mug produces richer motion data than a robot being steered to do the same thing, and people cost less than idle robot fleets. Researchers from Stanford, MIT, Georgia Tech, UC San Diego, ETH Zurich, and Meta have already worked with EgoVerse as a research project.

The money follows the pattern

Mecka says it crossed a $100 million annualized revenue run rate in June and expects $300 million by year end. Its customers are described only as several of the top frontier robotics labs and multiple of the world’s biggest tech companies, paying both for the data and for help wiring it in. The angel list reads like a tech-industry guest book: DoorDash CEO Tony Xu, former Snowflake CEO Frank Slootman, and Milan Kovac, who previously ran Tesla’s Optimus humanoid program.

The Financial Times reported that robotics companies and physical-AI labs have raised nearly $48 billion so far in 2026, much of it earmarked for the costly business of generating training data. Mecka is not alone in the hunt. XDOF is reportedly in Series B talks at a $1.2 billion valuation, while Scale AI and Micro1 are expanding from language-model data into robotics. Crowded markets usually mean the demand is real.

The quiet lesson

The pattern should feel familiar to anyone who watched the language-model boom. Scale AI, Mercor, and Surge built human-data supply chains that taught machines to talk. Mecka is running the same playbook for motion: if LLMs needed a million human annotators, humanoid robots may need a million human demonstrators. The model labs get the headlines and the chipmakers get the revenue, but the quiet money is flowing to the data layer underneath.

And the question for anyone building robots has changed. It is no longer whether you can afford the hardware. It is whether you can afford the ten thousand hours of human motion your robot still needs to watch.