One day after Nvidia posted the largest quarter any semiconductor company has ever reported, the machinery it built to keep the rest of the AI economy solvent went quiet.
On Wednesday, Aug. 26, Nvidia reported second-quarter revenue of $96.2 billion for the period ended July 26 — up 18% sequentially and 106% year over year — with data center revenue accounting for roughly $89 billion of it. Chief executive Jensen Huang guided to $108 billion for the current quarter and told analysts to expect 70% revenue growth in fiscal 2028.
By Thursday evening, The Wall Street Journal reported that Nvidia had put on hold a program it announced only eight weeks earlier: the AI Compute Partnership, a financing scheme under which Nvidia extended credit support to smaller AI cloud providers in exchange for a slice of the revenue those providers earned renting Nvidia chips. In the same quarterly filing that disclosed the record numbers, Nvidia said commitments under the program totaled $36 billion, with individual agreements typically running six years.
Shares in the so-called neoclouds fell Friday morning. CoreWeave, Nebius and IREN all traded lower, with IREN dropping more than 7% to roughly $38 — though that move was compounded by its own results, released late Wednesday, showing a quarterly loss of $0.41 per share on revenue of $137.23 million, down 27% from a year earlier.
What the program actually did
Announced July 1, the AI Compute Partnership was Nvidia's most explicit attempt yet to underwrite demand for its own silicon. The structure was unusual: Nvidia would provide credit backing — effectively a take-or-pay guarantee against unsold GPU capacity — and in return collect 50% of revenue above a preset base hourly rate once a provider's rental income cleared that threshold. Nvidia named Sharon AI and Firmus as the first two partners under the model.
The economics were designed to work twice. Nvidia booked the chip sale up front, then participated in the rental income stream those chips generated. For a capital-starved provider without an investment-grade balance sheet, the guarantee was the difference between a financeable data center and a slide deck.
What broke it, according to the Journal's reporting, was not the money. It was a control clause. Nvidia told some cloud providers they could rent the guaranteed chips only to Nvidia-approved customers, and signaled a preference for capacity spread across several smaller AI companies rather than leased wholesale to a single large one. A number of providers objected, arguing that choosing their own customers was the one prerogative they were not willing to sell. Separately, Nvidia employees raised concerns — internally and with current and prospective customers — that the arrangement invited antitrust scrutiny.
Nvidia has not conceded the framing. “The new business model we introduced in July that opens up compute access to the fast-growing AI ecosystem is still in place and continues to evolve due to high demand,” a company spokesperson said. It is a carefully built sentence: it affirms the model without addressing whether specific deals were shelved. Reports suggest Nvidia may rework the program or fold it into a different initiative.
Why this matters
The neocloud sector exists in the gap between what hyperscalers will build and what frontier labs need. It is financed almost entirely with debt, and that debt is priced off the assumption that the GPUs securing it will earn revenue for years. Nvidia's guarantees converted a speculative residual-value question into something a credit committee could underwrite.
Strip that out and the cost of capital moves. CoreWeave raised an $8.5 billion syndicated loan in March backed by its contract with Meta; a separate 2025 agreement obligates Nvidia to purchase CoreWeave's unsold capacity under a deal initially valued at $6.3 billion, running through April 2032. That backstop is distinct from the paused program and remains in force. But the signal matters more than any single contract: the sector's implicit lender of last resort has just demonstrated that its support is revocable, and revocable fast.
There is a second reading, and it is less bearish. Analysts have spent months warning about circular financing — a chipmaker extending its customers the means to buy its chips, booking the sale as revenue, and carrying the credit risk off to the side. If Nvidia is retreating from that structure because its own lawyers flagged antitrust exposure and its own investors flagged concentrated liability, that is a company reducing risk in a market where demand does not require the subsidy. Huang's own line on the earnings call — “compute is revenue. And demand is accelerating” — argues the guarantees were never load-bearing to begin with.
Wall Street has not capitulated. CoreWeave, Nebius and IREN all hold Moderate Buy consensus ratings. Citizens analyst Gregory Miller reaffirmed a Buy on IREN, noting that its “underlying AI cloud business continues to gain momentum,” while Bernstein's Gautam Chhugani said the company is “guiding for an exponential scale-up” toward $500 million in annual recurring revenue. The average IREN price target of $77.44 implies roughly 91% upside.
The tension is that both readings can be true. Nvidia can be simultaneously the healthiest company in the market and the entity whose withdrawal reprices an entire tier of its customers. The July program was an admission that some buyers could not finance themselves on their own credit. Pausing it does not make that problem disappear; it relocates it.
What to watch
Three things. First, whether Nvidia restructures the program or quietly lets it lapse — the difference between a legal cleanup and a demand signal. Second, the terms on the next neocloud debt raise: credit spreads will price the missing guarantee faster and more honestly than any equity analyst will. Third, whether a regulator picks up the thread Nvidia's own employees pulled. A company that pauses a $36 billion program over antitrust worries has, in effect, drafted the first paragraph of someone else's inquiry.
“The new business model we introduced in July that opens up compute access to the fast-growing AI ecosystem is still in place and continues to evolve due to high demand.”— Nvidia spokesperson, Statement to media, Aug. 28, 2026