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 allocation to a single humanoid robotics company. The deal provides Figure with priority access to next-generation inference chips optimized for low-latency robotics workloads, a configuration that Nscale typically reserves for hyperscaler contracts. Figure declined to comment on the equity percentage or board representation.
The commitment consists of two tranches: $1.8 billion in infrastructure deployed by Q3 2025 at Nscale's Reno facility, and an additional $1.7 billion contingent on Figure meeting undisclosed production milestones by Q1 2026. The Vera Rubin architecture runs inference workloads at 40% lower latency than the H100 clusters currently powering Figure's Gen 2 humanoid prototypes, according to internal benchmarks shared with limited partners. Nscale's CFO confirmed the company financed the commitment through a combination of balance sheet cash and a $2.1 billion credit facility arranged by Morgan Stanley.
The deal changes the unit economics of humanoid deployment. Figure's current cost structure requires $12,000 per robot per year in cloud inference fees at AWS rates. The dedicated Nscale infrastructure drops that figure to an estimated $3,200 annually, assuming Figure reaches 50,000 units deployed by 2027. That margin shift makes automotive and logistics partnerships economically viable at price points 30-40% below what Boston Dynamics and Tesla quote for comparable form factors. Three automotive OEMs and two warehouse operators are in advanced negotiations with Figure for pilot deployments beginning Q4 2025, contingent on the Nscale infrastructure coming online as scheduled.
The transaction also signals Nscale's pivot from fungible cloud capacity toward application-specific infrastructure plays. The company previously avoided equity stakes in customer companies, preferring to remain a neutral infrastructure layer. Taking a position in Figure suggests Nscale sees humanoid robotics as a distinct compute category with economics that justify vertical integration. Goldman Sachs estimates the humanoid inference market will reach $18 billion annually by 2030, with 70% of that spend concentrated in logistics, manufacturing, and automotive use cases. Nscale is reportedly in discussions with two other robotics companies for similar infrastructure-plus-equity arrangements, with term sheets expected by March.
Allocators should track Figure's production milestones in Q4 2025, which determine whether the second $1.7 billion tranche deploys. Watch for public statements from automotive OEMs regarding humanoid pilot programs, which will validate or undermine the unit economics thesis. Nscale's credit facility includes covenants tied to Figure's revenue growth, details of which remain confidential but typically involve quarterly revenue thresholds beginning 18 months post-signing.
The first Vera Rubin clusters go live in Reno in 127 days. Figure's Gen 3 humanoid, which the infrastructure is designed to support, begins limited production in 143 days.