Figure AI

The BMW Spartanburg pilot is real and the results are meaningful. One Figure 02 unit worked in the sheet-metal department, handling 90,000+ parts across 1,250 operational hours. The facility produces BMW X3 vehicles; Figure handled stamped sheet-metal parts that require consistent grip and orientation.

The numbers are honest on their face: 90,000 parts sounds like a fleet deployment. One robot over eleven months, working a fraction of production capacity, is a sophisticated proof-of-concept. BMW produced 30,000+ X3 vehicles during the same period using conventional automation and human workers.

What the pilot actually proved: Figure 02 can work in a live industrial environment, handle real production parts, and run for 1,250 hours without a reliability-ending failure. That is the threshold evidence that justified bringing Figure 03 into Hall 52 for iX5 logistics — the next step up in operational scope.

Figure 03's wireless inductive charging (2 kW) and 10 Gbps mmWave data link are designed for exactly this environment — continuous factory operation without manual cable connections.

Figure ended its OpenAI partnership in February 2025 and moved to a proprietary model called Helix — a Vision-Language-Action (VLA) model that runs entirely on-device. This is architecturally significant: it means Figure robots do not require a cloud connection to reason and act.

Helix uses a dual-system architecture. High-level reasoning (planning, object identification, task sequencing) runs at 7-9 Hz. Motor control (joint torques, grip force, balance) runs at 200 Hz. The separation allows the reasoning layer to update slowly while the body reacts in real time.

The OpenAI split was commercial as much as technical — Figure was paying for API calls on every robot interaction. On-device inference eliminates per-query costs at scale and removes a single point of failure. For a factory environment running 24/7, both matter.

Figure does not sell robots through a public channel. BotQ, the dedicated production facility, is targeting 12,000 units per year. As of April 2026, the line was producing one robot per hour — up from one per day earlier in the year. That ramp suggests enterprise availability is a near-term commercial goal, not a distant roadmap item.

The BMW relationship and the TIME recognition give Figure the commercial credibility to justify enterprise commitments. The $39B valuation reflects investor conviction that BotQ can scale. What it does not reflect is a consumer product in the near term — the per-unit cost at current production volumes puts Figure outside the range of individual buyers.