Nscale & Figure AI Compute | Scaling Humanoid Robotics

Infrastructure for Physical Intelligence

Nscale has entered a multi-year partnership with Figure to provide high-performance compute infrastructure for humanoid robotics. The agreement centers on the deployment of the NVIDIA Vera Rubin platform, which will support Figure’s next-generation Helix models. The move represents a significant commitment to hardware resources, with initial plans involving up to 100,000 NVIDIA GPUs. Figure will gain a dedicated compute provider, while Nscale secures a position as a strategic shareholder in the robotics firm.

Financial details of the arrangement point to a substantial investment. The parties have committed $3.5 billion in compute resources for the initial phase. This figure is expected to grow, with the companies aiming to scale the project to more than $6 billion over the coming years. This capital injection underscores the intense demand for training capacity in the race to achieve functional physical intelligence in machines.

Deployment Timeline and Geographic Footprint

The infrastructure work will begin with deployments in Barstow, Texas. Operations are scheduled to launch in the second half of 2027. By concentrating massive GPU clusters in this location, the companies aim to reduce latency and improve the throughput required for training advanced robotics models. Barstow serves as the hub for this initial build-out.

Integrating the power and orchestration layers of Nscale with the hardware stack provided by NVIDIA is a core component of this strategy. Figure intends to use this setup to manage its models across every stage of the lifecycle. This spans the initial training phases through to final deployment in real-world environments. The vertical integration of these assets aims to keep performance metrics high while maintaining efficiency throughout the training cycle.

Wider Industry Implications

The robotics and AI industry is currently navigating a period of massive infrastructure growth. Companies building humanoid hardware need more than just mechanical design. They require vast amounts of GPU power to process the visual and spatial data necessary for robots to navigate complex, unstructured environments. Partnerships like the one between Nscale and Figure indicate that the bottleneck for robotics is increasingly becoming a question of data center availability and computing scale rather than just motor control or sensor integration.

This trend toward specialized compute-for-robotics ventures reflects a shift in how firms view their operational overhead. By offloading the burden of managing massive GPU farms to partners like Nscale, Figure can focus its resources on its core competency of robotics software and hardware design. This strategy mirrors other industrial partnerships where specialized infrastructure providers support firms working on heavy computing tasks. As the demand for humanoid robots in manufacturing and logistics continues to rise, the need for this dedicated infrastructure will likely dictate which players successfully move from prototyping to mass deployment.