Paid Twice

The line that moved people was not the $96.2 billion.

NVIDIA (NVDA) reported fiscal Q2 on August 26. Revenue was $96.2 billion. Diluted EPS was $2.46. The stock still fell. That print is on our NVDA earnings page. The argument that followed is about a second check on the same silicon.

NVIDIA’s own blog, posted with the new AI-cloud model, says the structure lets the company earn “both standard product revenue and a share of the cloud revenue on the supported capacity.” The 10-Q, filed the same day as the call, is more careful and more useful: “If certain criteria are met, we will participate in revenue share generated by the AI clouds from third-party customers, which may contribute to revenue in the future.” The company also says those partners “procure our data center infrastructure products and we commit to cloud service agreements,” typically six years, totaling $36 billion as of July 26, 2026, and that the commitments “decrease as capacity is used by third-party customers or by us.”

On the August 26 call, Kress put it in one sentence: “Independent capital still underwrites every deal on its own merits. We're not making loans. In this model, we get paid twice, once on the hardware sale and again through the share of rental revenue, a highly reoccurring stream layered on top of a onetime equipment purchase.” Hear it on the NVDA earnings page.

Then came the pause.

Anissa Gardizy reported it on X at 2:53 p.m. on August 27, and the Journal screenshots in that post are the text we have. Headline: “Nvidia Pauses Revenue-Sharing Deals With AI Cloud Companies.” Dek: “The company is rethinking the program less than two months after announcing it.” The story, by Gardizy and Berber Jin, says NVIDIA “paused some deals in a new financing initiative that offered credit support to AI cloud providers in exchange for a share of revenue, according to people familiar with the matter.” It says some employees “expressed concern to current and potential customers that the program could draw antitrust scrutiny,” and that “there are sensitivities around the extent to which the chip giant can dictate how their customers do business.” It says NVIDIA “stepped back from the program last week,” that “the precise reason for the decision couldn’t be learned,” and that it “could revamp the program in the future or fold it into another initiative.” We did not open the article on wsj.com. Those sentences are from the screenshots she posted.

Ed Zitron wrote that Kress “literally said on the call last night that this was a way that NVIDIA would ‘get paid twice,’ despite having discontinued the program a week beforehand,” and that “they’re denying this report.” We have Kress’s sentence from the official transcript. We do not have a NVIDIA denial in a filing or a press release. The 10-Q does not mention a pause.

If the program is live, it is a second way to get paid on a rack NVIDIA already sold. If it is paused, the 10-Q still matters, because the larger number is not the revenue share. It is the guarantee.

In August 2026, after the quarter closed, NVIDIA entered guarantees “capped at a total of $105 billion” to support “a land, power, and shell buildout with affiliates of SB Energy Corp. (SB Energy) on behalf of a customer, an affiliate of OpenAI Group PBC (OpenAI).” The leases cover “approximately 4.25 gigawatts of IT load” at SB Energy’s PORTS Technology Campus in Pike County, Ohio. The 10-Q says each guarantee “generally becomes effective upon commencement of the applicable lease,” rises as each of nine phases is completed, “the first of which is expected in fiscal year 2029,” and is “expected to decrease over the course of each phase’s 20-year lease term.” Payment is triggered “upon certain tenant defaults.” The guarantees “are limited to defined portions of lease and power payments and not the full cost of the site or all of the tenant’s obligations.” They terminate if OpenAI gets a satisfactory credit rating, or when the lease is done. “In exchange for the guarantees, the site will exclusively host NVIDIA AI infrastructure, subject to limited exceptions.” NVIDIA also has an option, in its sole discretion, for about 3.8 additional gigawatts.

The table in the same note:

• Land, power, and shell guarantees for AI clouds: $3.5 billion
• SB Energy Corp. guarantees: $105.0 billion
• Total: $108.5 billion

That $108.5 billion is a maximum gross exposure in the filing. It is not cash out the door. It is not all due in 2026. The first phase is not even expected until fiscal 2029.

Exhibit 10.1 is the form of residual-value guaranty. The tenant name is redacted. The premises are in Piketon, Ohio. The exhibit itself states the total cap on this guaranty and related guaranties “shall not exceed $105,000,000,000,” tied to 4.25 GW of critical IT load. The covered-loss formula is redacted. So is the discount rate. So is the tenant. The 10-Q already named OpenAI. That is the “poor redaction” people are talking about. The exhibit does not add a second cap. It documents the first one.

A few numbers that are circulating are sloppy. Cash and cash equivalents on July 26 were $22.4 billion. That line is real. Stacking it against “$530 billion of obligations” is not a 10-Q total. The filing’s guarantee table is $108.5 billion. Separately, AI-cloud service commitments are $36 billion. Separately again, NVIDIA says it signed memorandums of understanding “to mobilize more than $500 billion of third-party capital over time,” and that “these and other preliminary arrangements may not lead to definitive agreements.” A $3.530 billion AI-cloud guarantee line in the derivatives table is $3.5 billion, not $530 billion. We are not adding those into one scare total.

The $1.5 billion equity in SB Energy is not in the 10-Q guarantee note. It is in NVIDIA’s August 17 press release: “NVIDIA will invest $1.5 billion in SB Energy, joining existing investors SoftBank Group and OpenAI.” The 10-Q option is “approximately 3.8 additional gigawatts.” The press release says 3.75. We use the 10-Q.

None of this is a buy or a sell. The earnings headline tells you what already shipped. The footnote tells you what NVIDIA is willing to stand behind so the next rack can be financed. Alerts exist for the second thing. The first thing you already know by the time the tape prints it.

Stay sharp.

Sources: NVIDIA 10-Q period ended July 26, 2026 (https://www.sec.gov/Archives/edgar/data/1045810/000104581026000075/nvda-20260726.htm); Exhibit 10.1 (https://www.sec.gov/Archives/edgar/data/1045810/000104581026000075/nvda2027q2ex101.htm); NVDA earnings page (https://pro.stockalarm.io/quote/NVDA/earnings); official IR earnings call August 26, 2026.