Mecka AI Is Turning Everyday Human Motion Into Robot Data
The robots may be artificial. The data teaching them how to move is very human.
Mecka AI, a young startup collecting real-world human movement data for robotics companies, is nearing a new financing round led by Sequoia Capital at a valuation of around $500 million, according to TechCrunch.
The proposed deal comes only about three months after Mecka announced a $60 million funding round led by Framework Ventures. The size of the new financing has not been disclosed, and TechCrunch reports that terms are still being negotiated and could change.
For a company founded in 2024, the valuation is striking.
The reason investors are interested is even more interesting.
Robots have a data problem.
The internet taught language models. Robots need the physical world
Large language models had access to an extraordinary training resource: decades of digital information.
- Webpages.
- Books.
- Code.
- Forums.
- Images.
Robotics doesn't have an equivalent dataset for physical intelligence.
A humanoid robot needs to understand how people manipulate objects, coordinate hands, move through environments and adjust when something doesn't behave exactly as expected.
That information exists everywhere in the physical world.
Very little of it has been structured into training data.
Mecka wants to change that.
The startup pays people to capture themselves performing ordinary tasks using smartphones and body sensors. That recorded movement can then be processed into training data for robotics systems.
Making coffee becomes data.
Repairing something becomes data.
Picking up and manipulating everyday objects becomes data.
Physical AI may create another Scale AI
The pitch sounds familiar because the AI industry has seen this movie before.
Scale AI and other data companies became valuable by providing structured human-generated data for machine-learning systems.
The next generation of robotics companies may require an equivalent infrastructure layer for physical behavior.
That means startups don't necessarily need to build humanoid robots themselves to benefit from the humanoid-robot boom.
They can sell the information needed to train them.
Mecka's founders recognized this despite not coming from traditional robotics backgrounds. The company was created around the view that scarce real-world interaction data could become a bottleneck for general-purpose robotics.
Investors are racing toward the bottleneck
Mecka isn't the only startup attracting money around this idea.
Other companies are collecting physical-world data through approaches including teleoperation, wearable sensors and first-person video.
The competition suggests investors increasingly view data collection as core robotics infrastructure rather than a temporary service.
That's important because robotics models have the same underlying problem language models once had.
Better models require more data.
But unlike internet text, high-quality physical-action data isn't sitting in one enormous public archive waiting to be scraped.
Someone has to create it.
Mecka is projecting software-like growth from a physical-data business
As of June, Mecka was projecting that it could finish 2026 at an annualized revenue run rate of around $100 million, according to reporting cited by TechCrunch.
If that trajectory holds, it helps explain investor enthusiasm.
The business sits in an unusual middle ground.
Its raw material comes from physical human activity.
Its customers are building advanced AI and robotic systems.
And the output is structured data that can potentially be sold at technology-company margins.
That combination is attractive during a period when billions of dollars are being invested into humanoids without a clear consensus on where all their training data will come from.
The startup doesn't need one robot maker to win
There is another appealing part of the model.
A robotics manufacturer is betting on its own hardware.
A data provider can potentially benefit from several manufacturers succeeding.
If multiple companies build warehouse robots, humanoids, industrial arms and home assistants, they may all need better real-world training data.
That gives Mecka the classic infrastructure startup pitch:
Don't bet on which application wins.
Sell something every winner needs.
What happens next?
The key challenge will be proving that massive quantities of captured human motion translate into better robots.
Not all data is equally useful.
Quality, diversity, labeling, sensor accuracy and the ability to convert movement into formats that models can learn from will determine which data providers become strategically valuable.
But investors are already signaling where they think the bottleneck is headed.
The next giant robotics company may build a humanoid.
Or it may build the dataset that teaches thousands of humanoids how to move.