What did Figure’s robots actually do at BMW?

Figure 02 performed a defined job at BMW’s Spartanburg plant: inserting sheet-metal parts for welding. BMW says it supported production of more than 30,000 X3 vehicles over ten months. Figure 03 moves the partnership into logistics, with parts sequencing and cart handling. Those are different machines, tasks and bodies of evidence. BMW’s June 25 account

Figure 02: a production contribution, with a specific scope

The vehicle figure tells us that the task was connected to a real production process. It is not a count of cars assembled entirely by humanoids, robots installed, or hours saved. Each of those would require a different measurement.

Figure’s November 19, 2025 report describes an eleven-month deployment and more than 1,250 operating hours. BMW uses a ten-month period for the production contribution. The descriptions are not interchangeable denominators. Dividing vehicles by months or hours to invent a robot productivity rate would require knowing exactly which machines, shifts and tasks each total covers. Figure’s deployment report

September 2026: the older fleet is mostly retired

This is a later status update, not the start of retirement. Figure had already announced the fleet-wide retirement and return of the BMW robots to headquarters in November 2025. The September publication does not date a new shutdown at BMW. It tells us the older generation is being phased out; it does not tell us how well its successor works for customers. The earlier retirement announcement

Figure 03: a new workflow needs new results

Figure’s June 30 account shows why the task is harder than repeating a fixed pick-and-place motion: the machine must handle parts while adjusting its body and moving a cart. BMW describes the intended sequence from unsorted components to a trolley and onward collection. These accounts explain the workflow; they do not provide a long-run service record for it. Figure 03 at BMW

Brett Adcock points to Figure 02’s experience as support for a flexible manufacturing workforce. The next useful test of that argument is whether the successor repeats its new job with acceptable supervision. A capable demonstration and an economical deployment can both matter without being the same achievement.

What would change the commercial assessment?

BMW’s September capital-market-day presentation places physical AI inside a broader platform strategy: reusable skills, production data and hardware chosen for each application. BMW describes scaling and cost-reduction goals across robot forms. That supports reading the partnership as a task-led development effort, rather than assuming every factory job needs a humanoid. The deck supplies no new Figure 03 throughput, uptime or earnings figures. BMW’s Physical AI workshop, slides 5–6

We would look for a defined operating period, tasks completed, interruptions and human interventions. Cost needs a boundary too: installation, maintenance and supervision can change the economics even if the robot performs its central motion well. Neither a vehicle total nor a video supplies that full calculation.

For readers assessing robotics-related investment themes, this is evidence of customer experimentation and a prior production task. It does not establish a material earnings benefit for BMW. An expanded, repeatable workflow would be stronger support for commercial adoption than applying the older model’s headline number to its successor.

The next comparison is Digit 5’s orders and delivery outlook: a customer commitment adds another kind of evidence, but it still has to become delivered, useful work. Tesla’s Optimus timeline begins at a different stage, with production and public availability needing separate checks.