London-based cloud computing company Nscale has agreed to supply humanoid robotics developer Figure with at least $3.5bn worth of computing capacity, and has also bought an undisclosed stake in the company. The arrangement, announced on 3 September, is among the clearest examples yet of a circular deal structure in which infrastructure providers and chipmakers become shareholders in the same customers they are paid to serve.

Key facts

• Nscale will provide Figure with at least $3.5bn of Nvidia Vera Rubin compute, with the intention to scale past $6bn.
• The deal covers up to 100,000 Nvidia GPUs, with the first systems arriving in the second half of 2027 at Nscale's Barstow, Texas site.
• Nscale has made an undisclosed strategic investment in Figure, becoming a shareholder and Figure's preferred compute provider for its Helix models.
• Nvidia is an investor in both Nscale and Figure, placing the chipmaker on both sides of a contract for its own hardware.
• Figure's Index programme collects about thirty minutes of human training video every second through a paid app.

A cloud provider investing in its own customer

Buying into a key customer is a playbook Nvidia has run for two years. Nvidia has built a roughly $99bn equity portfolio largely by taking positions in the AI labs and cloud providers that buy its chips. Nscale is following that playbook one layer down, with a robotics company rather than a model lab. The structure is tighter than a conventional customer-supplier relationship: Nvidia is an investor in Nscale and an investor in Figure, so the chipmaker has positions on both sides of a contract for its own hardware. The cloud provider in the middle now also owns equity in the customer at the end.

That kind of circularity has attracted scrutiny on Wall Street, where some analysts argue it can be used to create the appearance of demand that is larger than the underlying market. Nvidia has responded to similar concerns by describing its shareholdings as a way to accelerate adoption and share in the value created by its ecosystem. The Nscale-Figure deal extends that idea; rather than just equity in a chip buyer or a model lab, the London company has taken a stake in an end customer that will need a very large amount of compute to train and run its robots.

Nvidia calls it a flywheel

Jensen Huang has described the relationship with Figure as a robotics flywheel. Figure's models train on Vera Rubin hardware through Nscale's cloud. Nvidia's Isaac Sim validates the robot models. The finished models then deploy on Nvidia GPUs inside Figure's humanoid robots. That means Nvidia silicon appears at every stage of the process. The description is accurate, and so is the criticism that the loop is closed. Nvidia's finance chief has used the same language to defend its investments in frontier AI labs. The Nscale-Figure deal tightens that closed loop by adding an equity layer.

One line in the announcement goes further than most coverage has acknowledged. The two companies say they will explore scaling Nscale's supply chain with humanoids. Understood plainly, that means Figure robots could one day work in the data centres that deliver the compute used to train Figure robots. No timeline or financial target has been attached to that part of the plan, and it may remain an ambition. But it describes the end state both companies are arguing for. Nscale needs workers for sites in the Californian desert and the Norwegian Arctic. Figure needs somewhere its robots can do repetitive physical work at scale in a building controlled by its own customer.

Why Figure needs the compute

Brett Adcock, Figure's founder and chief executive, says the constraint is data and compute, not money. Helix, Figure's model family, improves the way any learned system improves: with more of both. Figure has made progress on the data side. Its Index programme collects about thirty minutes of human demonstration video every second, contributed by paid creators through a phone application. What the company has lacked is a place to run the training workload. Turning millions of hours of footage into robots capable of working a full shift requires sustained training runs at a scale Figure says it could not reach with its own infrastructure.

The partnership with Nscale gives Figure access to Nvidia's Vera Rubin platform when it arrives in late 2027. The first tranche of capacity is worth $3.5bn, and the companies have discussed expanding the agreement to more than $6bn. Figure is positioned as Nscale's preferred customer for physical-AI workloads, while Nscale becomes Figure's preferred compute provider for Helix.

Money has not been the brake on Figure's ambitions. In September 2025, the company closed a round of more than $1bn at a $39bn valuation. Parkway Venture Capital led the round, with Nvidia, Intel Capital, Salesforce and Qualcomm Ventures among the participants. The company's earlier $675m Series B in 2024 included Microsoft, Amazon's Industrial Innovation

Source:TNW | Investors-funding News