Figure AI identifies a scaling law for humanoid robots

By Splitfeed EditorialLast updated Oct 1, 2026, 15:20 UTC

Between September 29 and October 1, 2026, two notable developments emerged from Figure AI within days of each other. First, the company announced it had identified a scaling law for humanoid robots, a principle holding 56% of the time, suggesting that increasing compute or data in a structured way produces predictable improvements in robot capability. Then, by October 1, reports surfaced that Figure had deliberately decommissioned a fleet of its humanoid robots by sending them into a furnace in Finland, with the robots described as leaping into it, implying an active, choreographed end-of-life process rather than a passive disposal.

The scaling law claim matters because scaling laws were central to the rapid capability gains seen in large language models, and finding an analogous principle for physical robots would represent a meaningful shift in how the field approaches hardware and training investment. A law that holds 56% of the time is notable precisely because of that qualifier: it suggests a real but imperfect regularity, which raises immediate questions about what conditions produce the other 44% and whether the finding is robust enough to guide costly infrastructure decisions. The Finland furnace incident adds an unusual layer, as destroying a robot fleet in a dramatic fashion, whether for safety, data collection, testing structural limits, or simple decommissioning, points to a company moving through hardware generations quickly and with some deliberateness about how it retires them.

No reactions from the Pro-AI, Anti-AI, or Middle Ground camps have been published yet on these specific developments. Typically, Pro-AI voices would treat a robot scaling law as validation that the same compounding progress seen in software AI is now arriving in embodied systems. Anti-AI voices would likely focus on the 56% reliability figure as evidence that such claims are premature, and might point to the destroyed robot fleet as a sign of how much expensive trial and error still underlies the industry's confident announcements. Middle Ground commentators would probably ask for more methodological detail before drawing conclusions, noting that a scaling law with significant exceptions needs careful interpretation before it changes investment or development strategy.

The clearest next signal to watch would be a peer-reviewed or more detailed technical release from Figure AI explaining the conditions under which the scaling law holds, what it predicts specifically, and what drove the decision to decommission the Finland fleet. Those details would allow more grounded assessment of whether this represents a genuine structural insight or an early, qualified observation dressed in ambitious language.