Nscale is selling Figure $3.5bn of compute and buying a stake in it

Nscale agreed to sell Figure at least $3.5bn of computing power, under a deal covering up to 100,000 Nvidia GPUs running “Vera.” Nscale also bought a stake in Figure, suggesting a direct economic interest in the capacity and associated demand. Overall, this is a constructive partnership signal but the provided excerpt doesn’t quantify near-term revenue impact or margin implications.

Analysis

This is more narrative validation than fundamental delta for NVDA. The market will likely read it as another datapoint that AI compute demand is still being pre-funded and pre-committed, but the second-order read is more important: if the customer is taking an equity stake while also locking in capacity, part of the demand may be circular and less informative about end-demand than headline revenue suggests. That tends to support multiple durability for NVDA in the near term, but it does not yet prove a step-up in unit economics or utilization.

The real winners are the infrastructure intermediaries that can monetize scarce GPU supply without owning the full software risk: GPU lessors, data-center operators, and power-constrained hosting names. The risk is that these deals crowd out more price-transparent hyperscaler spend and create a later shakeout if utilization disappoints. For NVDA, the key incremental signal would be a visible tightening in lead times, ASP commentary, or a material revision higher in data-center capex guidance from multiple customers, not a single financing-backed placement.

Contrarian view: the consensus is treating every GPU allocation announcement as proof of unabated demand, but this may actually reflect capital-market engineering to keep the supply chain moving. Over 1-3 months, the stock can grind higher on sentiment alone; over 6-18 months, the thesis only holds if deployments convert into sustained inference workloads and repeat orders. What would falsify the bull case is evidence that these commitments are mostly balance-sheet supported, with delayed deployments, utilization below expectations, or any deceleration in NVDA’s data-center growth trajectory on the next two earnings prints.

Key Decisions for Investors

• Stay long NVDA, but treat this as a sentiment tailwind only: add on post-news weakness rather than chasing strength, since the incremental fundamental impact is modest.
• Use a pairs lens for the next 1-3 months: long NVDA / short a basket of AI infrastructure financing proxies with stretched expectations if the market starts rewarding any GPU-capacity announcement indiscriminately.
• Set an alert for NVDA next earnings and customer capex commentary: if data-center growth or backlog/lead-time commentary fails to reaccelerate, fade the move.
• Avoid paying up for near-dated upside purely on this headline; if positioning is already crowded, the better expression is stock ownership or a call spread only after confirmation in the next print.
• Watch for follow-on deals from other GPU buyers: if they also include equity stakes or vendor-financed structures, reduce confidence that the demand signal is organic.