Figure Secures 100,000 Nvidia GPUs in $3.5bn Deal to Scale Humanoid Robotics Training

Figure Commits $3.5bn for Nvidia GPUs Through Nscale Partnership

Figure has secured access to up to 100,000 Nvidia GPUs through a strategic partnership with AI compute provider Nscale, marking one of the largest infrastructure commitments in humanoid robotics1

1

. The initial $3.5 billion agreement, with plans to scale beyond $6 billion, will deploy computing power based on the Vera Rubin platform starting in the second half of 2027 at Nscale's Barstow, Texas facility2

2

. Nscale will also make an undisclosed strategic investment in Figure, becoming a shareholder and Figure's preferred compute provider1

1

.

The partnership addresses Figure's growing compute constraints as the company scales training for its Helix robotics foundation model. Brett Adcock, Figure's founder and CEO, emphasized that the company is increasingly limited by data and compute capacity rather than hardware manufacturing or funding2

2

. This infrastructure will support training AI systems that enable next-generation humanoid robots to perform real-world physical tasks across diverse environments.

Physical AI Flywheel Connects Training to Deployment

Jensen Huang, Nvidia's CEO, described the arrangement as activating a "robotics flywheel" where Figure's models train on Vera Rubin through Nscale's cloud, validate in Isaac Sim, and deploy on Nvidia GPUs inside Figure's robots1

1

. This physical AI flywheel creates a complete infrastructure pipeline for physical intelligence, positioning robotics as a new long-term demand driver for Nvidia's AI ecosystem beyond generative AI markets3

3

.

The partnership extends beyond compute provision. Both companies will explore using humanoid robots to support and scale Nscale's supply chain operations, potentially creating a scenario where Figure robots work in the data centres that train Figure robots1

1

. Nscale operates sites requiring staff in remote locations including the Californian desert and Norwegian Arctic, while Figure needs environments where its robots can perform repetitive physical work at scale.

Index Dataset Generates 30 Minutes of Training Data Every Second

Figure's Index dataset platform has surpassed 264,000 downloads across 108 countries with more than 44,000 weekly active users2

2

. Contributors upload enough video to generate 30 minutes of data every second, capturing the diversity and complexity of physical tasks needed to train general-purpose robotic intelligence2

2

. The platform spans household and workplace activities, covering hundreds of tasks, thousands of objects, and numerous environments.

Figure built its own data pipeline after conventional data suppliers failed to meet required scale, diversity, and quality standards. The system automatically filters incoming videos for technical, visual, and semantic quality before human reviewers check samples. Videos are processed for similarity, with duplicate or repetitive data removed to preserve dataset diversity2

2

. Figure has paid $15 million to contributors so far and plans to scale data and compute spending substantially as it moves humanoid robots from controlled demonstrations toward reliable agentic work.

Circular Investment Structure Mirrors Nvidia's Playbook

Nscale's equity investment in Figure replicates the structure Nvidia has employed for two years, where the chipmaker takes positions in labs and clouds that purchase its hardware1

1

. Nvidia is an investor in both Nscale and Figure, positioning the chipmaker on both sides of a contract for its own hardware while the AI compute provider in the middle now owns equity in the customer at the end1

1

. Some on Wall Street have flagged these circular arrangements as potentially making demand appear larger than actual market conditions.

Figure closed a funding round above $1 billion at a $39 billion valuation in September 2025, led by Parkway Venture Capital with Nvidia, Intel Capital, Salesforce and Qualcomm Ventures participating1

1

. A $675 million Series B in 2024 brought in Microsoft, Amazon's Industrial Innovation Fund and Jeff Bezos. Money has not been the constraint for Figure's development, with the company emphasizing that scaling physical intelligence requires immense compute rather than additional capital.

AI Infrastructure Demand Shifts Beyond Chatbots

The partnership signals that the next wave of AI infrastructure spending may come from teaching robots to understand and interact with the physical world rather than generate text3

3

. Humanoid robotics represents a fundamentally different AI challenge where systems must perceive the physical world, understand surroundings, and safely perform real-world tasks requiring continuous training on massive amounts of visual and behavioral data3

3

.

For investors who have largely viewed Nvidia's growth through hyperscalers and generative AI companies building large language models, Figure's commitment suggests another market is emerging. If humanoid robotics scales as companies envision, demand for AI infrastructure may increasingly come from training machines for physical intelligence. This could broaden Nvidia's customer base beyond cloud providers and AI labs, reinforcing the view that physical AI represents the industry's next frontier. Humanoid startups have raised $8.6 billion so far in 2026, though the sector is raising faster than it is shipping production systems1

1

. Nscale is preparing to go public, seeking up to $3 billion in a US IPO possibly as soon as September, having told prospective investors it holds around $51 billion in contracted revenue1

1

.