A major compute deal could give Figure the firepower it needs to push humanoid robotics closer to mass adoption
Humanoid robotics is moving into a new phase. Building capable robots is no longer just about better hardware. The bigger challenge is giving these machines enough data, training and computing power to learn how to operate in the real world.
That is where Figure’s latest partnership with Nscale comes in.
Figure and Nscale have announced a strategic partnership that could see up to 100,000 GPUs deployed on NVIDIA’s Vera Rubin platform, with the initial deployment targeted for the second half of 2027 in Barstow, Texas.
The agreement includes an initial $3.5 billion commitment for compute, with the potential to scale beyond $6 billion.
For Figure, this is much more than a large infrastructure deal. It is a bet that the next major step in artificial intelligence will come from machines that can physically interact with the world.
The compute race behind humanoid robots
Figure says it is entering a phase where its progress is increasingly constrained by two things: data and compute.
Its AI model, Helix, is designed to power increasingly capable humanoid robots. But training systems that can understand and act in the physical world requires enormous amounts of computing power.
Figure recently announced Index, an effort to build what it describes as the most diverse humanoid training dataset ever assembled.
The scale of data collection is striking.
- Index is generating 35 minutes of data every second
- The data is intended to help train Figure’s next generation of robotics models
- More data needs to be matched with significantly more compute
- The goal is to improve the ability of robots to understand and perform tasks in real-world environments
The basic idea is simple: more useful data gives the models more to learn from, while more compute allows that data to be processed and turned into better models.
For a company trying to build general-purpose humanoid robots, both are becoming critical resources.
Up to 100,000 GPUs changes the scale
The headline number in the deal is hard to miss.
Up to 100,000 GPUs.
The initial commitment represents $3.5 billion of compute, while the companies intend to potentially expand the agreement beyond $6 billion.
The initial deployment is expected to begin in the second half of 2027 at a facility in Barstow, Texas.
This gives Figure something it needs as its robotics models become more demanding: a much larger and more predictable supply of computing capacity.
Instead of treating compute as a supporting piece of the robotics business, Figure is effectively making it part of the core infrastructure required to develop its technology.
NVIDIA is becoming part of the robotics loop
The partnership also highlights how NVIDIA’s technology is becoming deeply connected to the development of physical AI.
The companies describe a process that connects several stages:
- Training: Figure trains its AI models using NVIDIA Vera Rubin through Nscale’s AI cloud
- Simulation: The models can be tested and developed using NVIDIA Isaac Sim
- Deployment: The resulting AI capabilities can ultimately run on NVIDIA GPUs inside Figure’s robots
- Real-world data: Robots generate new data that can be used to improve future models
This creates a continuous feedback loop.
The better the robots become, the more useful data they can generate. That data can then be used to train better models, which can lead to more capable robots.
That cycle is becoming one of the most important ideas in physical AI.
Nscale is getting more than a customer
The relationship also goes beyond Figure buying computing capacity from Nscale.
Nscale is making a strategic investment in Figure as part of the agreement.
The two companies will also explore whether Figure’s humanoid robots could eventually help scale Nscale’s supply chain.
That is an interesting part of the partnership because it connects two different sides of the AI infrastructure industry.
Nscale is building infrastructure to provide the computing power needed for AI. Figure is developing robots that could eventually perform physical tasks in environments built around human workers.
If humanoids become capable enough, they could potentially become part of the infrastructure and logistics ecosystem that supports AI itself.
That possibility is still developing, but the partnership shows how closely AI infrastructure and robotics are beginning to overlap.
Why this matters for Figure
Figure’s long-term ambition is much bigger than building a robot that can perform a handful of tasks.
The company wants to develop general-purpose humanoids capable of operating in environments designed for humans.
That requires robots to deal with situations that are difficult to pre-program individually.
A robot working in a home, warehouse or factory needs to recognize objects, understand instructions, respond to changes and adapt to situations it has not encountered before.
AI models such as Helix are intended to provide that intelligence.
But developing those models requires enormous amounts of training.
This is why the compute agreement is strategically important. Figure is trying to build the infrastructure required to scale the intelligence of its robots, not simply manufacture more machines.
The bigger bet on physical AI
For years, most of the AI boom has been focused on digital environments.
AI systems write code, generate images, analyze documents, answer questions and operate software.
Humanoid robots take AI into a very different environment.
They have to deal with:
- Physical objects
- Human movement
- Unpredictable surroundings
- Different lighting and layouts
- Fine motor control
- Real-time decision-making
- Safety around people
A model that performs well on a screen is not automatically capable of performing a physical task.
That is why physical AI could become one of the next major areas of competition in the technology industry.
Companies developing these systems need a combination of advanced AI models, huge datasets, simulation environments, specialized hardware and massive computing infrastructure.
Figure’s partnership with Nscale brings several of these pieces together.
What could happen next?
The initial deployment is not expected until the second half of 2027, so the full impact of the agreement will take time to materialize.
The more immediate takeaway is the scale of the commitment.
$3.5 billion initially, potentially more than $6 billion, and up to 100,000 GPUs.
That tells us how much computing companies expect to need if they want to build increasingly sophisticated physical AI systems.
For Figure, the priority is clear: collect more data, train larger and more capable models, and use those models to improve humanoid robots.
For Nscale, the partnership provides a major customer and strategic relationship in the rapidly growing physical AI market.
For NVIDIA, it is another example of its technology being positioned across the entire AI development cycle, from training and simulation to deployment.
The bigger question for investors
The interesting question is no longer simply whether humanoid robots will exist.
They already do.
The bigger question is how quickly they can become useful, reliable and affordable enough to operate at scale.
That will depend on several factors:
- How quickly AI models improve
- How much useful training data can be collected
- Whether computing costs fall as efficiency improves
- How reliably robots can perform real-world tasks
- How quickly manufacturers and consumers adopt humanoids
- Whether the economics make sense compared with human labor and existing automation
Figure is clearly betting that access to massive computing resources can accelerate that journey.
And with Nscale, NVIDIA and Figure now connected across computing, simulation, training and robotics, the pieces of a much larger physical AI ecosystem are starting to come together.
The race to build useful humanoid robots is increasingly becoming a race to build the AI infrastructure behind them.