Figure AI announced on 3 September 2026 a partnership with Nscale to deploy up to 100,000 Nvidia GPUs on the Vera Rubin platform, committing an initial $3.5 billion of compute with intent to scale to over $6 billion. The deal, targeting initial deployment in the second half of 2027 at Nscale’s Barstow, Texas data center, represents one of the largest infrastructure pledges by a robotics startup relative to capital raised. Public reports have put Figure’s capital raised at about $1.9 billion across earlier rounds, a total the Nscale announcement does not restate.
The compute will train Helix, Figure’s physical AI foundation model that powers the Figure 03 humanoid robot and the coming Figure 04. CEO Brett Adcock said the company is now largely bound by data and compute, noting that Figure’s Index dataset effort generates 35 minutes of training data every second. Nscale becomes Figure’s preferred compute provider and is making a strategic investment in the company, though the size was not disclosed.
Quick Take
- What happened: Figure AI committed $3.5 billion to Nscale for AI compute, with intent to scale to $6 billion.
- The hardware: Up to 100,000 Nvidia GPUs on the Vera Rubin platform, targeted for deployment starting second half 2027.
- The location: Nscale’s data center in Barstow, Texas, a site previously leased from bitcoin miner Ionic Digital.
- The context: Figure has raised $1.9 billion total, making this commitment nearly double its capital raised.
- The investment: Nscale is also making a strategic investment in Figure, size undisclosed.
In plain English
Figure AI just promised to spend $3.5 billion, possibly up to $6 billion, on cloud computing from Nscale to train the AI brain that runs its humanoid robots. That is a huge bet for a company that has raised only $1.9 billion so far. The deal covers up to 100,000 Nvidia GPUs that will start coming online in late 2027 at a Texas data center. This is a multi-year commitment, not cash changing hands today, but it shows how much compute power physical AI companies believe they will need to make robots that can work in the real world.
What happened
Figure announced the partnership on 3 September 2026, framing it as the compute runway needed to train the next generation of general robotics models. The company said it is entering a phase where it is largely bound by data and compute for training Helix, its physical AI foundation model. Last week Figure had announced Index, an effort to build what it calls the most diverse humanoid training dataset ever assembled, generating 35 minutes of data every second.
The deal targets initial deployment of the Nvidia Vera Rubin platform starting in the second half of 2027 in Barstow, Texas. Nscale CEO Josh Payne said physical intelligence is becoming AI’s next frontier, and the partnership will enable the future of AI robotics. Nvidia CEO Jensen Huang described the arrangement as activating the robotics flywheel: training Figure’s models on Nvidia Vera Rubin through Nscale’s AI cloud, validating them in Nvidia Isaac Sim, and deploying them on Nvidia GPUs in Figure’s robots.
Figure has raised $1.9 billion across four rounds, across previously disclosed rounds: a seed round, a $70 million Series A, a $675 million Series B, and just over $1 billion last September at a $39 billion valuation. The company has never disclosed revenue and has reportedly produced somewhere north of 1,000 of its Figure 03 robots. The new commitment is nearly double the total capital the company has raised.
The deal structure
the announcement noted that no money moved this week and that the $3.5 billion figure represents a multi-year commitment rather than a completed order. He questioned whether Figure can pay $3.5 billion from operations and pointed to circular-economy dynamics, noting that Nvidia has invested in both Figure and Nscale, and that Nscale may pursue an IPO. The $3.5 billion may describe rental spend over a term rather than purchasing 100,000 GPUs outright, Koetsier wrote, adding that Morgan Stanley rack-price math suggests $3.5 billion would buy far fewer than 100,000 Vera Rubin GPUs if purchased.
The Barstow site is leased by Nscale from Ionic Digital, a bitcoin-mining operation. Microsoft also has a large commitment at the same location. Koetsier raised questions about site capacity and noted that Texas has experienced interconnection pauses that could affect deployment timelines. The parties also said they will explore the potential to scale Nscale’s supply chain with humanoids, suggesting Figure robots could eventually be deployed in data-center operations.
The deal follows a broader pattern of AI companies making large infrastructure commitments to secure compute capacity. Anthropic, for example, secured $3.4 billion in Caterpillar gas generators to power its AI operations. Physical AI companies face particularly intense compute demands because they must train models on continuous streams of real-world sensor data, not just text or images.
Why it matters
The commitment size relative to capital raised signals how capital-intensive the race to build general-purpose humanoid robots has become. Figure is betting that scaling Helix will require compute on a scale previously associated with large language model labs, not robotics startups. The company’s Index dataset generates 35 minutes of training data every second, a rate that demands massive parallel processing to turn into useful model updates.
The partnership also highlights the emerging physical AI ecosystem around Nvidia’s platforms. Jensen Huang’s description of the robotics flywheel, training on Vera Rubin, validating in Isaac Sim, and deploying on Nvidia GPUs in robots, outlines a vertically integrated stack that Nvidia is positioning as the standard for the industry. Figure’s commitment locks the company into that ecosystem for the next several years.
The deal also raises questions about the business model for humanoid robotics. Figure has not disclosed revenue, and producing more than 1,000 robots at an estimated cost of hundreds of thousands of dollars each does not generate the cash flow to fund a $3.5 billion compute commitment. The company will need either rapid revenue growth, additional capital raises, or favorable financing terms from Nscale to meet the obligation. The strategic investment Nscale is making in Figure may be part of that financing structure.
What’s still open
- How Figure will finance the $3.5 billion commitment given it has raised only $1.9 billion and has not disclosed revenue.
- Whether the Barstow site can deliver 100,000 GPUs on schedule given capacity questions and Texas grid interconnection issues.
- The size of Nscale’s strategic investment in Figure and whether it offsets part of the compute commitment.
- How the $3.5 billion figure breaks down between rental fees over time versus capital expenditure for hardware.
- Whether Figure’s robots will actually be deployed in Nscale’s supply chain and data centers as the parties explore.
About the analysis
Researched and written with AI assistance under the editorial responsibility of the i10X News team. The title image is an AI-generated image. Read our AI Use Policy. Spotted an error? Request a correction.
This independent i10x analysis is based on Figure’s announcement of its Nscale compute partnership. Source names are given in the text; we do not link out to those pages.
i10x Perspective
This deal is either visionary or reckless depending on whether Figure can generate the revenue to pay for it. Committing nearly double your total capital raised to a multi-year compute contract before you have disclosed any revenue is a bet that Helix will unlock a step-change in robot capability that justifies the spend. The circular investment dynamics, Nvidia backing both parties and Nscale investing in Figure, suggest the deal may be structured more like vendor financing than a pure arms-length transaction. If Figure succeeds, this will look like the compute commitment that made general-purpose humanoid robots possible. If it stumbles, it will be a case study in infrastructure overcommitment by a pre-revenue hardware startup.






