Scale AI Buys Figure AI’s Data Team — Robotics Training Data Is Consolidating Fast
Scale AI acquired Figure AI’s data infrastructure division, claiming ~60% of the robotics training data market as Figure pivots to pure hardware development.
Scale AI just picked up Figure AI’s entire data infrastructure division in a deal that, price aside, says a lot about where the robotics AI industry is heading. The acquisition — announced by Scale AI with the financial terms undisclosed — hands Scale a dominant position in robotics training datasets and leaves Figure free to do what its founders actually want to do: build humanoid robots.
Figure AI, the humanoid robotics startup founded by Brett Adcock, had been operating both a hardware business and a data infrastructure arm. That’s a tough combination to sustain when you’re also trying to compete with Boston Dynamics and a growing field of well-funded robot companies. Selling the data operation wasn’t a retreat — it was a focus call.
What Scale AI Actually Gets
Scale AI already dominated enterprise AI data labeling before this deal. Add Figure’s robotics-specific data infrastructure and Scale now claims roughly 60% of the robotics training dataset market. That’s not a niche slice — robotics AI training data is one of the fastest-growing segments in AI infrastructure, as companies race to train autonomous systems on real-world physical interaction data.
The Figure data team brings specialized expertise in capturing, labeling, and structuring the kind of sensor-rich, multi-modal data that robotic systems require. Camera feeds, depth sensors, force feedback, motion capture — labeling this stuff accurately is harder than tagging images for a vision model, and it’s exactly the gap Scale needed to close at scale.
Figure’s Strategic Bet on Pure Hardware
Figure AI raised a $2.6 billion Series B in September 2024, which gave it the runway to make exactly this kind of strategic call. When you’ve got that kind of capital and a mandate to commercialize humanoid robots, running a data services operation on the side is a distraction. The sale lets Figure’s engineering team concentrate entirely on robot design, manufacturing, and deployment — the hard physical problem that no one has fully solved yet.
This mirrors a pattern playing out across the robotics industry: hardware companies are increasingly offloading data and software infrastructure to specialists, the same way cloud infrastructure got consolidated into AWS, Azure, and Google Cloud a decade ago. The companies that tried to own the full stack found the infrastructure work consumed resources that should have gone into the product.
What This Means for Robotics Startups
If you’re building a robotics startup right now and you need training data, your options just narrowed. Scale AI at 60% market share is the kind of concentration that makes buyers nervous and makes pricing conversations very one-sided. Smaller data labeling vendors still exist, but few have the depth in robotics-specific annotation that Scale now holds.
The optimistic read: Scale’s infrastructure is genuinely good, and having one dominant, well-resourced provider beats a fragmented market of mediocre ones. The pessimistic read: concentrated infrastructure suppliers have historically not been shy about using that position when contract renewal time rolls around.
Either way, the consolidation is real. Scale AI began as an image labeling startup and has methodically acquired or outcompeted its way into becoming the de facto backbone of AI training data across autonomous vehicles, drones, and now humanoid robotics. This deal is the latest step in a longer strategy, not a one-off.
What’s Next
Figure AI’s next milestone will be its first commercial deployment of humanoid robots — a crowded race that includes 1X, Apptronik, Physical Intelligence, and Tesla’s Optimus program. Shedding the data division removes overhead and sharpens the focus, but the hardware problem remains brutally hard. The question isn’t whether Figure made the right call selling the data team — it probably did — but whether the robotics market consolidates around two or three hardware players the way the data infrastructure market just did around Scale.
For Scale, the acquisition cements a position that will be very hard to dislodge. Robotics training data is sticky: once a company builds its pipelines and annotation workflows on your platform, switching is expensive. Sixty percent market share today tends to become seventy percent next year, not fifty.
