Robot data startup Mecka AI nabs $60M from Sequoia

Mecka AI has raised $60 million in a Series B funding round. Sequoia led the investment. Nvidia, Microsoft’s venture fund M12, and other investors joined. The round gives Mecka more financial support as it builds a business around robotics data.

Mecka collects human motion data for training humanoid and other robots. Its approach resembles the role that data companies play for large language models. Those companies provide human-generated examples that AI systems can learn from. Mecka pays people to record everyday activities while wearing body sensors and using smartphones.

TechCrunch previously reported that Mecka was nearing funding at a $500 million valuation. The new investors connect Mecka with major technology and artificial-intelligence companies. The round also reflects growing interest in real-world data for robots, alongside competitors such as XDOF and newer robotics efforts from Scale AI and Micro1.

Mecka AI is a startup founded in 2024 that collects and analyzes human motion data. It focuses on training humanoid robots and other kinds of robots. The company is building a data supply for robotics, similar to how Scale AI, Mercor, and Surge supply human-generated data for language-model systems.

Collecting human motion data means recording how people move while completing real activities. Mecka pays people to perform tasks such as making coffee or fixing cars. They use body sensors and smartphones during those recordings. The company can then analyze the captured movements as training material for robots.

This matters because robots need examples of actions in the physical world. Mecka’s work connects ordinary human behavior with robot learning. The article places the startup in a growing field that includes XDOF, Scale AI, and Micro1, all linked to collecting data for robot training.

Mecka AI’s Series B investment totals $60 million. That is the amount investors committed in the new funding round. Sequoia led the round, with Nvidia, Microsoft’s venture fund M12, and others participating. The money supports Mecka’s work collecting and analyzing data for robots.

TechCrunch had previously reported that Mecka was nearing a new round at a $500 million valuation. A valuation estimates what the entire company is worth in that financing context. It is different from the amount of money raised. Here, the reported company value is much larger than the new investment itself.

These figures place Mecka among startups attracting major attention in robotics data. The article also names XDOF, which was reportedly in talks for a Series B at a $1.2 billion valuation. That comparison shows investor interest extends beyond Mecka, although the two reported valuations are not the same.

Mecka gathers motion data by paying people to perform everyday tasks. The workers record themselves while completing activities in the physical world. This gives Mecka examples based on actual human movement rather than only written instructions or simulated actions.

The article gives two examples: making coffee and fixing cars. During these tasks, participants wear body sensors and use smartphones. The sensors and phones capture information about what people do as they move through each activity. Mecka then collects and analyzes that material for robotics training.

This process turns ordinary work into structured training data. The article does not specify the exact sensors, recording software, or analysis methods Mecka uses. It does establish the central method: paid participants perform tasks, body sensors and smartphones record them, and Mecka uses the resulting human motion data to support robot development.

Human task recordings can show robots how physical activities unfold. A robot-training system can use examples of human movement as data for learning useful actions. This is important for humanoid robots, which are intended to operate in environments designed around human tasks and movements.

Mecka pays people to record activities such as making coffee or fixing cars. They wear body sensors and use smartphones. Those tools capture the participants’ movements while the tasks happen. Mecka collects and analyzes the recordings, creating data that can help train humanoid and other robots.

The article does not describe a specific robot, software model, or completed capability. It does show the key mechanism: real human demonstrations become examples for robot training. As companies gather more real-world data, robotics developers may have more material for teaching robots physical behavior. Scale AI, Micro1, and XDOF are also expanding or operating in this broader data-for-robotics field.

Scale AI and Micro1 began with human-data platforms connected to language models. The article says they are also expanding into robotics. That move reflects a broader overlap between AI training needs: language systems learn from human-generated examples, while robots need examples of people moving and completing physical tasks.

Mecka illustrates the robotics version of this model. It pays people to record activities such as making coffee or fixing cars. Participants use body sensors and smartphones. The resulting motion data can be collected and analyzed for robot training. This gives robotics companies a way to obtain real-world examples instead of relying only on other forms of data.

The article does not detail Scale AI’s or Micro1’s specific robotics products. It does identify both as human-data platforms expanding beyond language-model work. Their move places them alongside Mecka and XDOF in a growing market for data that supports the development of humanoid and other robots.

Machine learning is a method in which a computer system learns patterns from examples. Instead of programming every response directly, developers provide data and a process for adjusting the system. The system uses those examples to make predictions or choose actions. This general explanation comes from established AI knowledge, not details stated in the article.

Human-generated examples help connect AI behavior to real tasks and preferences. For language models, people provide text or judgments. For robots, people can demonstrate physical activities. Mecka’s participants record themselves making coffee or fixing cars while wearing body sensors and using smartphones. Those recordings provide motion examples.

Large datasets can expose a system to more situations and variations, though data quality also matters. The article compares Mecka with companies supplying human-generated data for language models. It presents Mecka’s role as building a similar data foundation for humanoid and other robots, using real human movement as training material.