Figure Retired Its Robot Fleet at BMW Spartanburg After 30,000 Cars

Eleven months of work, more than 30,000 BMW X3 vehicles, and over 90,000 sheet metal parts loaded. Then Figure switched the fleet off.

Figure AI has retired its F.02 humanoid robots from BMW Group Plant Spartanburg in South Carolina. What began as a pilot finished looking a lot more like a shift pattern. Across the deployment the fleet accrued over 1,250 hours of system runtime, and by the tenth month the robots were completing full 10-hour shifts on the factory floor rather than short supervised windows.

What the Figure robot BMW Spartanburg run proved

The vehicle count is the number everyone quotes, but the runtime figure is the one that matters. Humanoid demos are cheap and easy to stage. A robot that stands in an assembly hall loading sheet metal for ten hours, day after day, for the better part of a year is a different claim, and the cars it worked on went to customers.

Figure has been open about the wear. The retired units came back bruised, which is roughly what you would expect from hardware that spent eleven months inside an active plant instead of a demo space. That honesty is more useful than a clean press photo, because it puts a maintenance cost against the labour saving. Nobody has published that ratio yet, and it is the number that decides whether any of this scales past a single customer.

Figure 03 is already in the building

The successor is not waiting on a funding round. Figure 03 has arrived in Hall 52, one of the assembly and logistics halls at Spartanburg, working on complex sequencing applications rather than the single repetitive task the F.02 fleet handled. BMW describes it as a commercial deployment following the pilot, which is a meaningful change in language from where this started.

Sequencing is harder than loading. It means picking the right part, in the right order, for a vehicle already moving down a line configured to one customer's order. Getting that wrong does not just stall a robot, it stalls a car. It also needs far more onboard inference than a fixed pick-and-place routine, which is why the energy cost of running AI workloads is starting to show up in factory planning documents rather than only data centre ones.

What to watch next

Whether Figure 03 hits comparable runtime numbers in a job with more decision points, because that is the gap between a robot that repeats and a robot that chooses. Models that look reliable in a benchmark still behave unpredictably once they are acting on their own in a live environment, and a factory floor is about as live as it gets.

The second signal is duller and more telling. If BMW extends to a second hall or a second plant, this stops being a project. One hall is an experiment run alongside the normal line. Two is a decision about how the plant gets staffed, and that decision would arrive with far less noise than the billion-user milestones that dominate AI coverage.