Mecka AI Raises $60M Series B to Build Robot Training Data Platform

Mecka AI has landed a $60 million Series B round to do for robots what data-labeling companies have done for large language models. The startup, founded in 2024, secured the funding from Sequoia Capital alongside backing from Nvidia, Microsoft’s M12 venture fund, and others, at a $500 million valuation.

The premise is straightforward but ambitious: humans get paid to perform everyday tasks while wearing body sensors and holding smartphones. These recordings become the training data that teaches humanoid robots and other robotic systems how to move, manipulate objects, and understand human behavior. It’s a labor-intensive approach, but one that mirrors the success of companies like Scale AI and Mercor in the LLM space.

The Data Economy Extends to Robotics

The robotics industry faces a fundamental challenge. Unlike AI language models trained on internet text, robots need embodied data - real humans performing real-world tasks. Mecka’s model capitalizes on this gap. By paying individuals to record themselves making coffee, fixing cars, or conducting repairs, the startup builds a dataset that teaches robots physical dexterity and contextual awareness.

This isn’t a novel concept anymore, but the capital flowing into it certainly is. The space is becoming increasingly crowded. XDOF, another robotics data collection startup, was reportedly in talks to raise its Series B at a $1.2 billion valuation. Meanwhile, established platforms like Scale AI and Micro1 are expanding their portfolios beyond language models into robotics training data.

A Proven Playbook

Mecka’s strategy mirrors what’s worked in adjacent markets. Scale AI built a multi-billion dollar business by providing curated training data for AI systems. Surge and Mercor followed similar paths. These companies proved that there’s significant value in the infrastructure layer of AI development - the unglamorous but essential work of data collection, labeling, and quality assurance.

Robotics presents an even more compelling opportunity. The sector is experiencing genuine momentum, with billions in venture capital flowing toward humanoid robots, autonomous systems, and industrial automation. But none of this works without quality training data. Mecka positions itself as the essential middleman, connecting the humans willing to perform repetitive motions on camera with companies desperate to train robots.

Market Timing and Competition

The timing matters here. Humanoid robots are transitioning from research curiosities to something approaching commercial viability. Companies like Tesla, Boston Dynamics, and Figure AI are racing to deploy these systems. That urgency creates demand for training data, and demand creates investment opportunities.

What makes this interesting is that robotics data differs fundamentally from text or image data. It’s harder to generate synthetically, harder to standardize, and requires real human participation. That creates a moat. While AI models can sometimes generate their own training data through synthetic means, robots need real-world motion capture to learn effectively.

The Sequoia Seal

Sequoia’s participation signals something important about the market’s perception of this opportunity. The venture capital firm has been deliberate about robotics investments, which suggests they see genuine scarcity and defensibility in what Mecka is building. The presence of Microsoft’s venture fund and Nvidia - a company with deep stakes in robotics infrastructure - reinforces the idea that this isn’t speculative.

What remains uncertain is whether Mecka can scale data collection as fast as robot manufacturers scale production. And whether the humans performing these recorded tasks will remain willing participants as the work becomes more standardized and commodified. The economics might work great for the startup, but the long-term sustainability of paying people to record themselves doing everyday tasks deserves scrutiny.

The real question isn’t whether robotics needs training data. It clearly does. The question is whether any single platform can become the dominant provider, or whether this market fragments into dozens of specialized data providers, each serving different robot manufacturers and use cases. History suggests the latter is more likely.

Source: TechCrunch