Figure AI has fundamentally shifted its bottleneck from building better robot bodies to acquiring enough computing power to train smarter robot brains. On September 3, 2026, the humanoid robotics company announced a multi-year strategic partnership with AI infrastructure provider Nscale to deploy up to 100,000 NVIDIA Vera Rubin GPUs (graphics processing units), with an initial compute commitment of $3.5 billion and plans to scale beyond $6 billion over time.

This deal marks a watershed moment in physical artificial intelligence. Rather than announcing a breakthrough in actuators, balance algorithms, or mechanical design, Figure is essentially saying: we have built robots that work, and now we need a data center the size of a small city to make them intelligent enough for mass deployment.

What Changed in Figure's Strategy?

For years, humanoid robotics companies focused on the hardware puzzle: how do you build a machine that can walk, balance, and manipulate objects with human-like dexterity? Figure solved much of that problem. The company's Figure 03 humanoid has reached production scale, with the company manufacturing over 350 units and ramping production from one robot per day to one per hour as of April 2026.

But Figure's leadership realized that hardware commoditization is inevitable. The real constraint now is artificial intelligence. The company's proprietary AI model, called Helix, powers all of Figure's robots across three hardware generations. To make Helix capable enough to operate autonomously in homes and factories, Figure needs massive amounts of training data and the computational horsepower to process it.

"Figure is entering a phase where we are largely bound by data and compute needed to train Helix. Our AI model, Helix, becomes more capable the same way every learned system does: with more data and compute. Today we're excited to partner with Nscale to bring the compute required to make this vision a reality," said Brett Adcock, Founder and CEO of Figure.

Brett Adcock, Founder and CEO, Figure AI

This shift reflects a broader realization across the robotics industry: you can engineer a robot that mimics human movement, but teaching it to understand context, adapt to new environments, and solve novel problems requires the kind of machine learning infrastructure that only hyperscale companies have historically built.

How Will Figure Use 100,000 GPUs?

The Vera Rubin platform represents NVIDIA's next-generation rack-scale AI system, introduced in January 2026. According to NVIDIA, Rubin delivers up to a 10x reduction in inference token cost and a 4x reduction in the number of GPUs needed to train mixture-of-experts models compared with the previous Blackwell platform. In practical terms, this means Figure can train more sophisticated AI models faster and more cheaply than before.

Figure recently launched an initiative called Index, which crowdsources real-world video from tens of thousands of global contributors. The company stated that Index is generating approximately 35 minutes of training data every second, capturing human interaction footage that can be converted into motor policies for robots. That firehose of unstructured video requires enormous computational resources to process, label, and use for training.

The deployment timeline is important: initial GPU rollout is scheduled to begin in the second half of 2027 at Nscale's facility in Barstow, Texas. This means Figure's immediate training runs will still depend on existing GPU clusters, but by late 2027, the company will have access to one of the largest dedicated AI compute clusters for robotics in the world.

Why Does This Deal Matter Beyond Figure?

The Nscale partnership reveals how the robotics industry is maturing. Early-stage humanoid companies focused on proving that bipedal robots could work at all. Now that proof exists, the competitive advantage shifts to whoever can train the most capable AI models fastest. This requires capital, infrastructure, and long-term compute commitments that only well-funded startups can afford.

Nscale is also making a strategic equity investment in Figure and becoming the company's preferred compute provider. This vertical integration means Figure has guaranteed access to reliable computing resources, a critical advantage in an industry where GPU availability is often constrained. Additionally, both companies are exploring the potential to deploy Figure's humanoid robots within Nscale's own data center supply chains, creating a feedback loop where robots help operate the infrastructure that trains them.

NVIDIA's role in this ecosystem is equally significant. The company's founder and CEO Jensen Huang described the partnership as activating a "robotics flywheel": training Figure's models on Vera Rubin clusters via Nscale's infrastructure, validating physical locomotion and manipulation in NVIDIA's Isaac Sim simulation environment, and deploying the optimized policies onto NVIDIA GPUs embedded in Figure's robots.

Steps to Understanding the Compute-First Robotics Era

• Training Data Scale: Figure's Index initiative generates 35 minutes of training video every second, requiring massive compute to convert raw footage into actionable motor policies for robots.
• Hardware Commoditization: Building a humanoid robot that walks and balances is no longer the bottleneck; the constraint is now training AI models intelligent enough to operate autonomously in unstructured environments.
• Infrastructure as Competitive Advantage: Companies that secure long-term access to next-generation GPU clusters, like Figure through Nscale, gain a structural advantage in developing more capable AI models faster than competitors.
• Vertical Integration: Nscale's strategic investment in Figure and role as preferred compute provider creates alignment between infrastructure provider and robotics company, reducing supply chain risk.

What Are the Practical Implications?

The $3.5 billion initial commitment, with plans to scale beyond $6 billion, underscores that humanoid robotics is no longer a scrappy startup endeavor. It requires the kind of capital and infrastructure typically associated with large language model (LLM) development. An LLM is a type of artificial intelligence trained on vast amounts of text to predict and generate human language.

For Figure, this deal buys runway. The company raised over $1 billion in committed capital at a $39 billion post-money valuation in its Series C funding round in September 2025, with NVIDIA among the investors. The Nscale partnership ensures that capital can be deployed toward AI training rather than building out proprietary data center infrastructure, a significant operational advantage.

However, the timeline also reveals the long developmental horizon for general-purpose physical agents. Deployment doesn't begin until the second half of 2027, meaning Figure's most ambitious AI training runs are still years away. Building the power generation, cooling capacity, and physical infrastructure to support 100,000 GPUs in Barstow, Texas will itself be a multi-year engineering challenge.

The broader implication is clear: humanoid robotics has entered the era of hyperscale computing. Success will increasingly depend on access to cutting-edge GPUs, reliable power, and the capital to sustain multi-billion-dollar infrastructure commitments. For Figure, the Nscale partnership signals confidence that the company's hardware and AI architecture are sound, and that the remaining challenge is simply scale.