The GPU neocloud becomes Figure’s preferred compute provider — and takes equity in the customer it will bill billions. The financing structure of the model layer has arrived in physical AI.
What Happened
Per their joint announcement on PR Newswire, Nscale — a GPU cloud provider competing in the crowded neocloud tier — and Figure, the humanoid-robotics company behind the Helix foundation model, have signed a multi-year strategic partnership in which Nscale becomes Figure’s preferred compute provider. The arrangement is structured around a committed floor of at least roughly $3.5 billion, with stated intent to scale past approximately $6 billion — though that upper figure is scaling language, not a signed obligation, and headline numbers in AI infrastructure deals routinely exceed what ultimately gets spent. Hardware specifics: up to roughly 100,000 of NVIDIA’s next-generation Vera Rubin GPUs at a site in Barstow, Texas, with initial deployment targeted for the second half of 2027 and dedicated to training and running Figure’s Helix robotics models.
The detail that makes this more than a large purchase order: Nscale is also making a strategic equity investment in Figure. The amount is undisclosed — no ownership percentage can be stated, and none should be inferred. Two things are therefore simultaneously true: this is an officially announced, jointly confirmed deal with a concrete multi-billion-dollar floor, and its most structurally significant element — the supplier taking equity in the customer — carries numbers that are either uncapped (the intent-to-scale figure) or undisclosed (the equity). Both facts belong in the same sentence.
What is confirmed cleanly: a GPU neocloud and a humanoid-robotics company entered a multi-billion-dollar, multi-year compute agreement in which the compute supplier also became a shareholder in the company it will bill. That structure — not the headline number — is the durable part of this announcement.
The key insight: When the company selling the compute also owns equity in the company paying for it, the demand signal embedded in the deal is partly self-funded. That makes it a softer indicator of independent market demand than an arm’s-length contract — not necessarily unsound, but worth naming plainly before reading it as evidence that physical AI has arrived at commercial scale.
The Structural Read
The model layer of the AI stack has run on a form of circular financing for the better part of two years. Chipmakers hold equity in GPU clouds. GPU clouds hold equity in labs. Labs transact with the clouds that hold their equity, and the clouds transact with the chipmakers whose chips they resell. Revenue and investment portfolios point at the same counterparties, and the money moves in a loop — flattering every participant’s growth metrics while making it genuinely difficult to read any single deal as clean, independent demand. The Nscale–Figure partnership imports that structure, intact, into humanoid robotics.
The mechanics are straightforward: Nscale commits to supply Figure compute worth billions over multiple years. Nscale simultaneously takes equity in Figure. Some portion of the future revenue Nscale will book from Figure is therefore underwritten, indirectly, by Nscale’s own capital sitting inside its customer. This is what circular financing looks like when it crosses a layer boundary. It is not necessarily unsound — anchor-tenant infrastructure has been built this way for decades, and a young company like Figure could not plausibly finance a 100,000-GPU buildout on its own balance sheet. The point is not impropriety. The point is that a compute commitment where the supplier is also a shareholder is a different, structurally softer demand signal than an arm’s-length contract, and conflating the two overstates what the deal tells us about independent market pull for physical AI.
BE Framework — Circular / Loop Financing
Supplier Equity in Customer: The Backstop Economy Extends
When a supplier takes equity in the customer it will bill, the deal simultaneously creates revenue (for the supplier) and a capital obligation (from the supplier’s own balance sheet back into the customer). The loop means that part of the supplier’s future revenue is underwritten by the supplier’s own prior investment — making the gross demand figure partially self-funded. This pattern, well-established in the AI model layer, now has a foothold in embodied AI. When the money starts moving in loops through a new layer, it is a sign that layer has become investable enough to attract financial engineering — and early enough that customers cannot finance buildouts independently.
Read from each side, the logic is clean on its own terms. For Nscale, a neocloud competing against hyperscalers with vastly larger balance sheets, the playbook is to lock a marquee, long-duration anchor tenant in a frontier category before the big clouds do — and capture equity upside on top of the compute margin. That turns a supply contract into a strategic position: Nscale earns revenue from Figure’s compute spend, and participates in Figure’s valuation appreciation if physical AI scales. The risk Nscale is taking on is that Figure scales; the reward is that it holds a piece of the company when it does.
For Figure, the deal solves the actual constraint on scaling Helix. Embodied AI hits the same compute wall as every other frontier model program — it just points the output at a robot instead of a text response. Securing priority access to up to roughly 100,000 next-generation Vera Rubin GPUs, years in advance, at a dedicated site, is the kind of infrastructure position that a humanoid robotics company at Figure’s stage cannot easily replicate on the spot market. The compute access is real and the prioritization is genuinely valuable, regardless of the financing structure behind it.
The larger significance is the pattern, not the deal. The financing techniques of the AI compute boom — multi-year compute commitments bundled with equity, vendor stakes, and intent-to-scale headline numbers — are now being applied to the robotics layer. Physical AI is being capitalized on the same circular, cross-held terms as the model layer that preceded it. That is a signal about where the industry’s financial center of gravity is moving, not a guarantee that the underlying bets pay off.
The Neocloud Anchor-Tenant Playbook
“A compute commitment where the supplier is also a shareholder is a different, softer kind of demand signal than an arm’s-length contract. Worth naming plainly — because reading it as the latter, when it is the former, overstates what the deal tells us about physical AI’s commercial readiness.”
Three Implications
IMPLICATION 1 — The Neocloud Differentiation Strategy
Neoclouds cannot outspend hyperscalers on general infrastructure. Nscale’s move — anchor tenant plus equity in a frontier category — is the credible alternative: lock a marquee customer early, turn the supply relationship into a portfolio position, and compete on strategic alignment rather than rack count. If Figure scales, Nscale holds a piece of it. If physical AI becomes the next battleground for compute, Nscale has a multi-year head start with a named partner. This is the playbook other neoclouds will study.
IMPLICATION 2 — The Compute Wall Is Embodied AI’s Real Constraint
The bottleneck for humanoid robotics foundation models is identical to the bottleneck for language models: access to next-generation GPU capacity at scale, secured in advance. Figure’s Helix program requires the same training infrastructure as any frontier model lab — it just outputs robot behavior policies instead of tokens. Deals like this one will determine which physical AI programs can actually scale their training runs in 2027 and which will be capacity-constrained. The compute access is the real prize, and the financing structure is the mechanism for getting it.
IMPLICATION 3 — Physical AI Now Has Circular Financing; Read Market Signals Accordingly
As these cross-holdings multiply across the physical AI layer — compute providers taking equity in robotics companies, robotics companies committing to compute providers — the aggregate revenue and investment figures in the sector will increasingly reflect money moving in loops rather than independent external demand. Analysts and observers reading deal flow as a signal of genuine commercial traction should weight arm’s-length demand separately from vendor-financed demand. The Nscale–Figure structure is not unique to this deal; it is the template the sector appears to be adopting.
