Mecka sells critical robot training data

Crossing $100M run rate that quickly shows Mecka is selling into an acute bottleneck, not a nice to have tool. Robotics labs need thousands of hours of people doing real tasks in real kitchens, warehouses, and work sites, then need that footage cleaned, reconstructed, and packaged so models can learn from it. Mecka bundled dataset sales, custom collection, and deployment work into one offering, which lets spending ramp fast when a few large labs decide data is the constraint on model progress.1 2 3 4

• This market behaves more like high end infrastructure services than SaaS. A customer can buy an existing dataset, or pay for a new capture program for a specific robot, task, and environment, so contract size can jump quickly once a lab moves from experiments to production scale collection.1 2
• The closest internal analogue is Scale, which expanded from generic labeling into physical AI data collection with global contributors, dedicated collection infrastructure, and integrations on production robots. Mecka is narrower and earlier, but the same pattern is visible, whoever owns hard to gather training data can become core to model development.3 4 5
• The important distinction is that Mecka is not building the robot. It is selling the picks and shovels around robot learning, capture hardware, human demonstrations, reconstruction models, and on site integration. That can make revenue appear unusually fast because multiple robot makers and large tech buyers can all spend on the same data layer.1 2 6

The next step is a shift from one off data projects to an always on data and deployment loop. As more robots move into commercial settings, the valuable supplier will be the one that can keep collecting new edge cases from live environments, turn them into training data quickly, and feed model improvements back into deployments.1 3 5