Nvidia's PORTS Support Adds a $105B AI Infrastructure Credit Backstop
Jensen Huang says small ecosystem investments can unlock far larger product sales. SB Energy's filings show the bargain in harder terms: $1.5 billion already prepaid under an equity-linked forward, another $1.5 billion committed at the IPO, and up to $105 billion of staged lease support tied to OpenAI.
Jensen Huang defended Nvidia's AI-infrastructure financing at Goldman Sachs on September 10, arguing that strategic capital can function as distribution for Nvidia architecture. His shorthand was blunt: "I put in one, and a hundred comes back." The PORTS-Pike relationship with SB Energy gives investors an unusually detailed public record against which to test that claim.
The capital is more complex than the August announcement suggested. A legal share-purchase agreement signed August 17 commits Nvidia to buy $1.5 billion of non-voting Class N stock in SB Energy's IPO vehicle at the IPO price. That purchase has not closed: the contract says payment occurs immediately after the IPO closes, subject to its conditions.
A separate August 17 prepaid-forward contract has already moved cash. It states that Nvidia prepaid $1.5 billion to Energy Global, LP, SB Energy's indirect parent, in exchange for an equity-linked claim that can settle into shares in a public listing. The two arrangements therefore should not be described as one already-funded $3 billion equity cheque to SB Energy Inc. One is funded at the parent level; the other is a future $1.5 billion purchase tied to the IPO.
The larger exposure is contingent support. Nvidia has entered residual-value guarantees covering leases for about 4.25GW of IT load at SB Energy's PORTS-Pike campus in Ohio, where OpenAI is the tenant. The aggregate guarantee is capped at $105 billion and becomes effective in phases as relevant facilities satisfy ready-for-service conditions. OpenAI will deploy Nvidia's full-stack DSX infrastructure at the covered capacity, subject to limited exceptions.
Nvidia is buying distribution with both capital and its balance sheet
Nvidia estimates that each hardware generation deployed at PORTS-Pike could represent roughly 1.5 million GPUs and $150 billion to $200 billion of Nvidia revenue. That is potential product revenue, not contracted profit, and it cannot be netted against the guarantee. It does explain why Huang is willing to use capital and credit to secure the site.
The $1.5 billion future share purchase is small beside one generation of potential hardware sales, while the prepaid forward puts another $1.5 billion of cash into the SB Energy ownership chain before the IPO. The guarantee supplies something different again: credit support rather than investment capital. Together they show that Nvidia's demand-creation strategy is no longer confined to minority equity stakes.
The guarantee also has a defined loss path. It is generally triggered by OpenAI insolvency that causes a lease default or by OpenAI's failure to make required lease payments. SB Energy can seek recovery through a replacement lease or asset sale, and Nvidia generally covers the shortfall between that recovery and the guaranteed minimum value. Nvidia can have remedies that include taking over a lease, requiring a re-lease or sale, or paying specified costs while recovery proceeds.
OpenAI has separately agreed to reimburse and indemnify Nvidia for amounts Nvidia actually pays to the landlord. That right is valuable when OpenAI remains solvent. It is weakest in the state that matters most to the guarantee: OpenAI has already failed to pay or become insolvent, while replacement demand and residual infrastructure values may also be under pressure.
The tail is correlated with Nvidia's core earnings
Nvidia already reports $3.5 billion of maximum gross exposure under land, power and shell guarantees for other AI-cloud partners. Adding PORTS takes disclosed maximum gross contractual guarantee exposure to $108.5 billion. That is a ceiling on contingent obligations, not debt outstanding or an expected loss.
The structure is economically attractive if OpenAI performs and the campus operates at high utilization. Nvidia gets exclusive infrastructure positioning across a huge site, the guarantee activates only as phases become ready for service, and exposure declines as lease payments are made. A small amount of direct capital can support a much larger installed base of Nvidia systems.
The risk appears in covariance. A broad AI-demand break could hit Nvidia accelerator sales, weaken the credit of a frontier-model customer and reduce what a replacement tenant would pay for the infrastructure at the same time. The contracts move part of that correlated tail from customers and landlords onto Nvidia's balance sheet.
Huang's defense therefore survives the simple charge that this financing automatically creates artificial revenue. PORTS can be rational distribution because the direct capital is small beside one generation of potential hardware sales. The stronger shareholder conclusion is that Nvidia is purchasing that distribution with more than equity: it is also underwriting customer and residual-value risk that becomes most expensive in the same downturn that would already pressure its semiconductor franchise.
Sources
Wall Street Journal, Huang on Nvidia's circular-financing criticism, September 10: https://www.wsj.com/livecoverage/stock-market-today-dow-sp-500-nasdaq-09-10-2026/card/-money-quote-jensen-huang-on-nvidia-s-circular-deals-yvOvWzMOfTOOoT7BNZTJ
SEC, Nvidia/SB Energy $1.5 billion share purchase agreement: https://www.sec.gov/Archives/edgar/data/2133037/000162828026059639/exhibit1028-sharepurchasea.htm
SEC, Nvidia/Energy Global $1.5 billion prepaid forward: https://www.sec.gov/Archives/edgar/data/2133037/000162828026059639/exhibit1027-prepaidforward.htm
Nvidia August 17 PORTS 8-K: https://www.sec.gov/Archives/edgar/data/1045810/000104581026000069/nvda-20260817.htm
Nvidia fiscal Q2 2027 filing and guarantee disclosures: https://www.sec.gov/Archives/edgar/data/1045810/000104581026000075/nvda-20260726.htm