Figure melted its old robots with Schwarzenegger watching — the stunt says more about moats than machines

The facts are short and strange. Figure, the humanoid robotics company, took its retired Figure 02 units to a steel furnace in Finland and melted them down. The stated motive, per the company, was protecting patented technology. Arnold Schwarzenegger attended the event, and the recovered steel is to be turned into limited-edition memorabilia.

Our thesis: the spectacle is marketing, but the policy underneath it is the signal. A company does not pay to incinerate working hardware unless it believes the hardware itself is the secret. In software-first AI, the moat is weights and data, and old versions get deprecated into oblivion at no cost. In embodied AI, the moat is actuators, joints, thermal design, hands — things an engineer can learn by holding them. Melting the fleet is an admission that Figure thinks its advantage can be reverse-engineered from a used machine.

The staging is an unmistakable nod to the Terminator ending, and casting Schwarzenegger makes the reference explicit. That is a choice worth naming for what it is: a robotics company borrowing the cultural imagery of killer machines because it converts cheaply into attention. It tells us nothing about whether the next generation works. Keep demonstrated capability and promotional theatre in separate columns — the furnace is a press event, not a benchmark.

There is a real cost on the other side of the ledger. Retired humanoids are scarce, expensive research instruments, and university labs and safety researchers are chronically short of them. Destroying rather than donating or dismantling them under agreement keeps the physical layer of this industry closed to independent scrutiny at exactly the moment outsiders most need to study how these machines fail. That is a governance problem dressed as an IP decision.

Our read: this is the humanoid sector entering its phone-upgrade phase — generations churning fast enough that a whole fleet becomes obsolete inventory, with competition fierce enough that secrecy beats salvage. Short term, expect more opacity, more spectacle, and more claims that outrun what the robots actually do. Long term we remain optimistic about where this goes: machines that take on dangerous, repetitive and physically punishing work, and that help cover the care shortfall ageing societies cannot staff. Getting there requires hardware that outside researchers can inspect, not only steel ingots sold as souvenirs.