Figure Signs $3.5B Nscale Deal for Up to 100,000 GPUs for Humanoid AI
The partnership gives Figure a path to far larger training capacity, but its first Texas deployment is still targeted for the second half of 2027.
Listen to this story
The audio brief
Story brief
3 key pointsFigure’s bottleneck is shifting from collecting robotics data to securing enough capacity to train on it. The deal’s practical test is whether Nscale can deliver a Vera Rubin installation in Barstow, Texas, during the second half of 2027. The initial $3.5 billion compute commitment could expand beyond $6 billion, while Nscale separately takes an undisclosed strategic stake. The capacity is intended to turn Figure’s...
- 01
Deployment is planned for Barstow, Texas, in the second half of 2027—not immediately.
- 02
The agreement targets up to 100,000 Vera Rubin GPUs, with scaling beyond $6 billion still only an intent.
- 03
Nscale’s investment in Figure is strategic, but its size was not disclosed.
Figure has signed a partnership with Nscale to deploy up to 100,000 GPUs on NVIDIA’s Vera Rubin platform for training Helix, its AI model for general robotics and humanoid robots. Initial deployment is targeted for Barstow, Texas, in the second half of 2027, putting the planned training buildout on a long timetable.
The agreement begins with a $3.5 billion compute commitment and includes an intent to scale beyond $6 billion. Nscale is also making a strategic investment in Figure; the companies did not disclose its size.
Compute is the binding input
Figure says it is largely constrained by the data and compute required to train Helix, and that data alone cannot solve the training problem. Its Index data-collection effort is generating 35 minutes of humanoid training data every second, according to the company. The Nscale agreement is intended to supply the compute side of that equation.
A system designed around the GPU
Vera Rubin is not simply a collection of graphics processors. NVIDIA says the platform is built through codesign across seven chips and five rack trays. The company said production was ramping at partners including CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius.
NVIDIA also said a CoreWeave benchmark running DeepSeek-R1 measured 10 times more throughput per megawatt than Grace Blackwell NVL72. That is a company-reported result from a separate workload, not a performance forecast for Figure’s robot training.
The planned loop reaches into deployment
NVIDIA describes the intended workflow as training Figure’s models on Vera Rubin through Nscale’s AI cloud, validating them in Isaac Sim, and deploying them on NVIDIA GPUs in Figure robots. It links the cloud training system, simulation software and hardware intended for the robots themselves.
The companies will also explore using Figure humanoids in Nscale’s data-center supply chain. That element remains exploratory, not a committed robot deployment. The concrete milestone is the planned Barstow installation, whose delivery will determine when Figure can begin using this new capacity.
Editorial analysis
Our Read
Our Read: Figure is putting a price on a specific bet: that more training data and more computation will yield more capable physical behavior. The agreement makes the test unusually concrete. Watch whether Nscale starts the Barstow deployment on its stated schedule, then whether Figure turns that capacity into a larger commitment and demonstrably stronger Helix models. The arrangement also places Nscale in two roles—as Figure’s compute provider and investor—while leaving open the possibility that Figure’s robots could eventually participate in Nscale’s own supply chain.
Sources
- figure.aiFigure and Nscale Sign Strategic Partnership For Up to 100,000 GPUs on the NVIDIA Vera Rubin Platform
- blogs.nvidia.comNVIDIA Vera Rubin Driving Performance Per Watt, Lowest Token Cost for Partners Worldwide
