Nscale Sells $3.5B Compute Stake, Buys Interest

Nscale has secured a massive deal to supply Figure with at least $3.5 billion worth of computing power, while simultaneously taking an equity stake in the robotics company. The agreement, announced on September 3 by the London-based infrastructure firm, represents a significant shift in how AI hardware and robotics companies are aligning their financial and operational interests.

The partnership involves access to up to 100,000 Nvidia Vera Rubin GPUs. These systems will be housed at Nscale’s facility in Barstow, Texas, with initial deployments scheduled for the second half of 2027. While the immediate commitment is valued at $3.5 billion, both companies have expressed an intent to scale this investment beyond $6 billion as demand grows.

Josh Payne, Nscale’s founder and chief executive, described the move as an entry into a new market segment. He stated that “Physical intelligence is AI’s next frontier,” arguing that Figure is pushing boundaries further than current inference and agentic work models allow. Alongside the compute agreement, Nscale made an undisclosed strategic investment in Figure, making it a shareholder and the preferred provider of cloud resources for Figure’s Helix models.

A Circular Investment Structure

The structure of this deal mirrors strategies previously employed by Nvidia, drawing comparisons from financial analysts. Bloomberg’s Lynn Doan noted that by investing in a key customer, Nscale is following a playbook Nvidia has utilized for two years. This approach has faced scrutiny on Wall Street, where some investors view such circular arrangements as methods to artificially inflate perceived demand.

Nvidia itself holds positions in both Nscale and Figure, effectively sitting on both sides of the contract for its own hardware. This creates a closed loop where the chipmaker benefits regardless of which entity generates revenue. Jensen Huang characterized this dynamic as the “robotics flywheel.” In this model, Figure’s models train on Vera Rubin through Nscale’s cloud infrastructure. Nvidia’s Isaac Sim software then validates these models before they are deployed onto Nvidia GPUs within Figure’s robots. This ensures that Nvidia silicon is utilized at every stage of development and operation.

One notable aspect of the announcement is the suggestion that Figure robots may eventually work within the data centers that train them. The companies stated they would explore scaling Nscale’s supply chain with humanoids. While no specific timeline or figures were provided, this concept addresses a mutual need: Nscale requires labor for its remote sites in the Californian desert and the Norwegian Arctic, while Figure needs environments where its robots can perform repetitive physical tasks at scale.

Figure’s Compute Constraints

Brett Adcock, Figure’s founder and chief executive, identified data and compute availability as the primary bottlenecks for his company, rather than capital. He explained that “Helix improves the way any learned system does, he said, with more of both.”

Figure has made strides in data collection through its Index program, which gathers thirty minutes of human training video every second via a mobile app used by paid creators. However, processing this vast amount of footage into functional robotic capabilities requires sustained computational power that Figure could not previously access.

Financial constraints have not hindered Figure’s growth. In September 2025, the company raised over $1 billion at a $39 billion valuation, led by Parkway Venture Capital with participation from Nvidia, Intel Capital, Salesforce, and Qualcomm Ventures. Earlier, in 2024, a $675 million Series B round included investments from Microsoft, Amazon’s Industrial Innovation Fund, and Jeff Bezos. Additionally, Adcock founded Hark, an AI hardware startup that raised $700 million at a $6 billion valuation in May.

Nscale’s Pre-IPO Positioning

Nscale is making these substantial commitments while preparing for a public listing. In March, the company closed a $2 billion Series C round at a $14.6 billion valuation. It subsequently secured approximately $3 billion in debt financing to support campuses in Texas and North Carolina. Reports indicate that Nscale is seeking up to $3 billion in a US IPO, potentially as early as September, and has disclosed around $51 billion in contracted revenue to prospective investors.

The company operates facilities in Narvik and Glomfjord in Norway, Loughton in Essex, and Texas, with partner capacity in Portugal, Iceland, and North Carolina. The Barstow site is designated specifically for the Figure workload. This deal adds a layer of complexity to Nscale’s backlog, as a portion of its contracted revenue now points toward a customer in which it holds equity.

Sector-Wide Funding Surge

The humanoid robotics sector is experiencing rapid capital inflows. So far in 2026, startups have raised $8.6 billion, nearly double the total for all of 2025. Skild AI secured $1.4 billion in January at a valuation exceeding $14 billion, backed by SoftBank, Nvidia, and Jeff Bezos. Apptronik raised past $935 million in February, while Physical Intelligence was reported to be in talks for roughly $1 billion at an $11 billion valuation in March.

Market projections suggest strong long-term demand. Barclays estimates that humanoids could offset 60% of China’s projected shortfall of 37 million workers by 2035. Nvidia continues to court various manufacturers to capitalize on this trend.

Despite the high valuations and funding rounds, the fundamental question remains whether these robots can generate enough economic value to justify their computational costs. Figure faces a $3.5 billion compute bill against a $39 billion valuation, with Nscale positioned to profit through both revenue recognition and equity appreciation. The true test for these companies will not be their next funding round, but rather the first commercial customer who purchases the robots based on their utility rather than their narrative.