Nscale and humanoid-robotics company Figure have signed a multi-year computing partnership with an initial commitment valued at $3.5 billion. The Next Web reported the agreement on September 6, 2026. Nscale’s September 3 announcement says the relationship could scale beyond $6 billion and potentially deploy up to 100,000 Nvidia GPUs.
A large compute reservation with a later start
The companies target the first GPU deployments for the second half of 2027 at a site in Barstow, Texas. The planned infrastructure uses Nvidia’s Vera Rubin platform to train and serve Figure’s Helix models. Nscale will become Figure’s preferred compute provider, covering power, cloud infrastructure and orchestration.
The distinction between a commitment and deployed capacity matters. Up to 100,000 GPUs is a ceiling in the announcement, not hardware already installed or workloads already running. The release does not provide a delivery schedule for the full amount, utilization levels, energy consumption, performance measurements or independent validation of the planned system.
Supplier and shareholder at the same time
Nscale is also making a strategic investment in Figure and will become a shareholder. That aligns the infrastructure provider with the customer’s growth, but it also makes the commercial relationship more complex. Revenue from compute and the value of the equity stake could both benefit if Figure expands.
For readers and investors, the arrangement should therefore be evaluated as more than a conventional cloud contract. The supplier has a financial interest in the robotics company, while Figure is concentrating an important part of its future training capacity with one preferred provider. The announcement does not disclose the size or terms of Nscale’s equity investment.
Why robotics demands so much infrastructure
Figure says its Helix models improve with more data and compute. Humanoid systems need training across perception, language, planning and physical action, as well as simulation and validation before deployment. Nscale says the workflow will combine Vera Rubin infrastructure, Nvidia Isaac Sim and GPUs embedded in Figure robots.
Those components describe an intended technology stack, not proof that general-purpose humanoids are ready for widespread home or industrial use. Hardware reliability, safety, data quality, task success and operating cost remain separate constraints. More compute may accelerate experimentation without resolving each of them.
What to watch next
The first meaningful checkpoints will be construction and power readiness in Barstow, delivery of the initial GPUs, disclosed capacity actually available to Figure and evidence that Helix workloads are using it. Clearer information on the equity investment and contract obligations would also help distinguish firm commitments from expansion options.
The agreement is significant because it links AI infrastructure financing, robotics model development and ownership in one transaction. Its eventual impact depends on physical deployment and measurable robotics progress, not the maximum numbers in the announcement.