Figure and Nscale have announced a planned partnership to provide access to up to 100,000 GPUs based on NVIDIA’s Vera Rubin platform for training Figure’s Helix robotics models. Initial deployment is targeted for the second half of 2027 in Barstow, Texas, meaning the announced capacity is not operational today.
Figure describes the agreement as an initial commitment of $3.5 billion of compute, with an intention to scale to more than $6 billion. Nscale is also making a strategic investment in Figure, but the companies have not disclosed the investment’s value or terms.
Contents
What Figure and Nscale announced
The partnership connects three parts of Figure’s planned robotics development pipeline: data generation through its Index effort, model training using Nscale’s AI cloud and NVIDIA’s Vera Rubin platform, and deployment of trained models on hardware in Figure robots.
Figure says its Index effort is generating 35 minutes of data every second and is intended to build a diverse dataset for training humanoid robots. The company says the data and computing power required to train Helix are becoming major constraints on its development.
The announcement also describes a possible industrial application beyond model development. Figure and Nscale will explore using humanoid robots to scale parts of Nscale’s supply chain. This is a stated possibility, not a commitment that the robots will be deployed in those operations.
NVIDIA CEO Jensen Huang described a workflow in which Figure’s models would be trained on Vera Rubin through Nscale’s AI cloud, validated in NVIDIA Isaac Sim and then deployed on NVIDIA GPUs in Figure robots. Isaac Sim is a simulation environment: it can allow robot systems to be tested in virtual settings before or alongside testing on physical machines.
Figure says the partnership is intended to provide the compute needed to train the next generation of AI models for general robotics. The company has also linked the arrangement to its ambition of bringing humanoid robots into homes worldwide, but the announcement does not provide a consumer release date, price, safety record or availability plan.
Why the compute matters for Helix
Training is the process of adjusting an AI model using data and computation so that it performs better at its intended tasks. Larger training runs generally require more accelerator hardware, memory, storage and networking.
The number of GPUs alone does not establish usable delivered capacity or model performance. Outcomes also depend on hardware availability, networking, memory, storage, software, data quality, utilisation and the training methods used.
Simulation can expose an AI-controlled robot to virtual environments before or alongside physical testing. This can increase the amount of training and evaluation performed without using physical hardware for every trial, although the announcement does not provide measurements showing how much simulation or physical testing will be used.
After training, inference is the process of using the model to generate outputs or control actions. Robotics inference may need to run on local hardware inside or near the robot because control decisions can be affected by connectivity and latency constraints. The proposed partnership therefore describes a pipeline from large-scale cloud training to simulation and then deployment on robot-mounted NVIDIA GPUs. It does not specify the exact robot hardware, model sizes, networking design or software versions involved.
The announced scale is significant as a planned infrastructure commitment because Figure is explicitly treating access to data and compute as limits on its robotics development. If the infrastructure is delivered as described, it could give the company substantially more capacity for training and evaluating models. The announcement does not yet show that additional compute will produce a particular level of robot capability.
What to watch
The most important milestones will be whether initial deployment begins in Barstow during the announced second-half-2027 period and how many GPUs are actually installed and made available to Figure.
Further evidence will also be needed on whether larger training runs produce measurable gains in Helix’s robotics performance, including reliability, safety, manipulation, navigation and the ability to generalise beyond training conditions.
The terms of Nscale’s strategic investment in Figure remain undisclosed. It is also not yet clear whether the two companies will proceed with using humanoid robots in Nscale’s supply chain, or what energy, cooling, networking, utilisation and operating costs would accompany infrastructure at the announced scale.