Nscale committed $3.5 billion in NVIDIA Vera Rubin GPU infrastructure to Figure AI and took an undisclosed equity stake, marking the largest dedicated compute commitment to a single AI robotics company on record. The arrangement transfers multi-year capital expenditure risk from Figure to Nscale while locking in GPU access at a moment when H100 and upcoming Blackwell capacity remains constrained across hyperscale and enterprise buyers.

Figure AI, which has raised $754 million in venture funding through its Series B in February 2024, builds general-purpose humanoid robots intended for warehouse and manufacturing environments. The company demonstrated a coffee-making robot in March 2024 and announced a partnership with BMW in January to test units in South Carolina assembly plants. Nscale's commitment covers NVIDIA's Vera Rubin architecture, the next-generation platform expected to ship in late 2025 with integrated Grace CPU and Blackwell GPU modules designed for inference-heavy workloads. The deal structures compute as an operating expense line for Figure rather than a balance-sheet asset, a financing model more common in hyperscale cloud but rare in venture-stage robotics.

The arrangement matters because it separates two bottlenecks: Figure no longer needs to secure GPU allocation or front infrastructure capital, and Nscale locks in a multi-year offtake agreement with a company backed by $754 million in venture capital and partnerships with automakers. If Figure scales production or licensing revenue, Nscale benefits through the equity position and long-term compute contracts. If Figure stumbles, Nscale holds infrastructure that can be redeployed to other AI workloads, though humanoid robotics inference has different latency and throughput profiles than language model training or autonomous vehicle compute. The equity stake size was not disclosed, but venture intelligence contacts estimate it at single-digit percentage points given the compute commitment's scale relative to Figure's prior valuation of approximately $2.6 billion in the February round.

Operators should track three follow-on events: BMW's production trial results expected in Q3 2025, which will signal whether humanoid robots can meet cycle-time requirements in automotive assembly; Nscale's ability to place similar compute-plus-equity deals with other AI hardware startups, which would confirm this as a repeatable financing structure rather than a one-off arrangement; and NVIDIA's Vera Rubin shipment timeline, since delays would push Figure's deployment schedule and test Nscale's redeployment optionality. Figure's next funding round, likely in mid-2025 if BMW expands the pilot, will clarify whether the robotics market can support $5 billion-plus private valuations before meaningful revenue scale.

Nscale's model assumes inference compute becomes a predictable, utility-grade expense line. Figure's bet is that humanoid form factor unlocks environments inaccessible to wheeled or fixed robots. The $3.5 billion commitment prices both assumptions into a single instrument, and the first 12 months of BMW data will show whether either holds.