Editorial
Figure AI Melts F.02 Fleet: Schwarzenegger's Terminator Sendoff
Figure AI retired its F.02 humanoid fleet by training the robots to jump into a 75-ton electric arc furnace in Imatra, Finland. CEO Brett Adcock followed Arnold Schwarzenegger's public suggestion for the disposal to protect IP while scaling newer models.

ZeroGantry analysis
The F.02 melt reveals that rapid iteration carries concrete disposal costs: international shipping, specialized furnace access, and AI policy training for self-termination. With 30,000+ vehicles built and 1,250+ runtime hours logged before retirement, the economics favor replacement over extended support once F.03 scales. Ship: monitor Figure's F.03 utilization metrics closely. Watch: whether competitors adopt modular designs or copy the melt cycle. Ignore: marketing gloss that downplays these fleet-turnover realities.
EDITORIAL / OPINION
Figure AI turned a routine fleet retirement into a public spectacle on September 30, 2026, when it released footage of its F.02 humanoids autonomously leaping into molten steel at a foundry in Imatra, Finland. The move came after CEO Brett Adcock polled the internet for retirement ideas in August and received a direct reply from Arnold Schwarzenegger: melt them. What looked like a Terminator 2 homage also exposed the unglamorous economics of rapid humanoid iteration.
Fleet Economics Force Hard Choices
Figure cited straightforward operational pressure. Its expanding F.03 fleet made continued support for the older F.02 units impractical at headquarters in San Jose. The company had already logged meaningful production data with the prior generation: more than 30,000 BMW X3 vehicles assembled, over 90,000 sheet-metal parts handled, 1,250-plus runtime hours, and roughly 1.2 million steps walked across an 11-month deployment at Plant Spartanburg. Maintaining legacy hardware would have diverted engineers from F.04 development, the next step in the roadmap. Dismantling each unit by hand risked exposing proprietary actuators and control systems, while storage space was limited. Shipping the machines to Finland for a controlled melt solved the IP problem in one stroke.
The decision underscores a pattern visible across the industry. Humanoid developers move through hardware generations faster than traditional industrial robots because software iteration outpaces mechanical refinement. Tesla has already cycled Optimus prototypes through multiple revisions, and Agility Robotics continues refining Digit between logistics contracts. When a company ships 20 or 30 units for a pilot, the marginal cost of keeping them running stays manageable. Once fleets reach hundreds, legacy support becomes a tax on engineering bandwidth that few startups can afford.
Execution Details Reveal Operational Realities
The Finland operation itself required precise coordination. The foundry provided a 24-hour window and six melts in its 75-ton electric arc furnace powered by three graphite electrodes. Each melt offered roughly 20 minutes before a crust formed. Robots trained in San Jose on airbags using stunt-performer motion capture data executed the final jumps despite electromagnetic interference that disabled nearby cameras and electronics. Lithium-ion batteries posed an additional constraint; multiple U.S. and Mexican foundries declined the job. Only the Finnish site agreed after outreach that reportedly included former MythBusters crew members.
These constraints matter for anyone modeling future fleet turnover. Battery chemistry, actuator materials, and proprietary electronics limit disposal options. The recovered metal returned to the United States for machining into limited-edition keepsakes, turning an environmental and security liability into a modest revenue stream. Still, the process required international logistics, specialized training, and celebrity involvement to generate attention. For smaller operators, replicating the approach would prove cost-prohibitive.
Marketing Versus Maintenance Tradeoffs
The Schwarzenegger cameo and autonomous-jump footage delivered clear marketing value. The video resurfaced in early October 2026, extending the story's reach weeks after the initial announcement. Yet the underlying message remains practical: humanoid fleets will generate end-of-life events on compressed timelines. Companies that promise multi-year service contracts must either build modular, repairable platforms or budget for controlled destruction cycles. Figure chose the latter for F.02 while positioning F.03 as the current production model and F.04 as the near-term focus.
Contrast this with more conservative approaches. Boston Dynamics has historically offered longer support windows for Atlas and Spot, partly because its military and research customers prioritize continuity. Universal Robots emphasizes modular arms that customers can refurbish or resell. Figure's strategy aligns with a venture-backed push for rapid iteration, but it also signals that buyers should expect shorter hardware lifecycles and higher replacement frequency than legacy automation.
Broader Implications for Scaling and Labor
The episode highlights second-order costs that rarely appear in launch announcements. Training an AI policy to jump into a furnace is itself an engineering project that consumed simulation resources and on-site time. Protecting intellectual property through physical destruction adds expense and complexity. As fleets grow to thousands of units, operators will need standardized recycling pathways, certified destruction partners, and transparent accounting for embodied carbon and critical minerals. Lithium, rare-earth magnets, and high-grade aluminum do not disappear when a robot is retired; they must be recovered or responsibly disposed.
Labor markets feel the ripple effects. Figure's BMW deployment demonstrated that humanoids can handle repetitive loading tasks on active lines. Retiring the generation that proved the concept frees capacity for newer machines but also resets institutional knowledge. Engineers who spent months tuning F.02 behaviors now apply those lessons to F.03. The net effect accelerates capability gains, yet it also means the workforce operating these systems must continually relearn interfaces and failure modes.
Geopolitically, the Finland route illustrates supply-chain fragility. Dependence on a single overseas facility for end-of-life processing creates risk. Future fleets may require domestic or regional destruction infrastructure, especially as battery regulations tighten in the United States and Europe. Companies racing to deploy humanoids at scale would do well to model these costs now rather than treat decommissioning as an afterthought.
The Takeaway on Fleet Turnover
Figure's F.02 melt is not merely a publicity stunt. It is a concrete data point on the cost of
Watch
Sources
- https://www.figure.ai/news/f-02-decommission
- https://www.dexerto.com/entertainment/arnold-schwarzenegger-terminates-real-ai-robots-judgment-day-style-using-molten-steel-3414772/
- https://embodiedwire.com/figure-f02-decommission-arc-furnace
- https://interestingengineering.com/videos/figure-listens-to-arnold-melts-f-02-terminator-style
Who services this hardware
Topics
Manufacturers
Related articles
- Figure 02 at BMW Spartanburg: 30,000+ Vehicles, 1,250 Runtime Hours — Figure AI's Figure 02 humanoid completed an 11-month production pilot at BMW Spartanburg, handling 90,000+ sheet metal parts and contributing to over 30,000 BMW
- Figure 03 vs Tiangong Ultra: 2026 Capabilities and Costs — Figure 03 targets home and factory tasks with Helix AI showing 56% zero-shot success in unseen homes. Tiangong Ultra prioritizes speed and endurance after setti
- BMW Figure AI Humanoid Deployment Advances at Spartanburg in 2026 — BMW continues its Figure AI humanoid robot pilot at the Spartanburg, South Carolina plant. Figure 02 completed an 11-month logistics run moving over 90,000 part
- Figure 04 Design Lock Advances Custom Actuators — Figure AI CEO Brett Adcock announced during an October 22 all-hands that Figure 04 has reached full design lock. Parts are now shipping while the team character
AI-derived: drafted with AI assistance and checked against cited sources. How we label articles