Nvidia pauses the rent-back guarantee that made small AI clouds bankable
Eight weeks after Nvidia started lending its balance sheet to small AI clouds, internal warnings about antitrust and about how much it was dictating to customers have stopped parts of the programme, and the operators who needed it most have no obvious substitute.
The Product Desk
What happened
- The Wall Street Journal reports Nvidia has halted some transactions under the financing programme it announced on 1 July, which extended credit to AI cloud buyers in return for a share of what the chips later earned.
- Employees inside Nvidia flagged that the structure could attract antitrust scrutiny and questioned how far the company could reasonably go in dictating how its customers run their businesses.
- Nvidia has not confirmed the pause, and a spokesperson said the July model remains in place and continues to evolve because of high demand.
- The two deals announced at launch were Sharon AI signing for up to 40,000 Grace Blackwell GB300 chips and Firmus taking commitments for up to 170,000.
Why it matters
- What it costs cost The cost falls on operators who cannot fund accelerators from their own balance sheet, who now pay for the missing guaranteed buyer either in a higher cost of debt or in a purchase they do not make.
- What doesn't add up contradiction One account says the revenue split was never disclosed while the other reports a 50 percent take above an undisclosed threshold, so a reader cannot yet tell whether the paused deals were expensive or merely restrictive.
- Who has to decide decision Anyone who wrote this credit line into a capacity plan has to re-underwrite that plan without knowing which version returns, because Nvidia has left both revising the terms and folding the scheme elsewhere on the table.
- What it sets up precedent If the programme returns with tenant approval intact, chip financing arrives carrying a say over who your customers are, and that clause belongs in the negotiation rather than the appendix.
What made the July programme bankable was the rent-back promise. Nvidia agreed to lease capacity a customer could not sell, which handed the lender at the end of the chain a guaranteed buyer and made borrowing against accelerators easier [c3b]. For an operator planning a 2026 buildout, that one sentence was the difference between a debt package and a hope. Nvidia designed the scheme precisely for buyers who could not fund chips outright, so pausing it removes the route for the companies with the fewest alternatives [15].
The price of that guarantee is where the two write-ups part company. TNW says the revenue-share percentages have never been disclosed, which is why outside analysis of the programme has been thin [27]. The Straits Times, carrying Reuters on the Journal's reporting, gives a figure: under the proposed deals Nvidia would take 50 percent of any revenue a provider earned through its chips above a certain threshold [28]. The threshold is missing, and without it 50 percent tells you nothing about the economics. Firmus said at launch it expected $25bn to $30bn in committed customer offtake over the first six years of its deployment [17], which averages $4.2bn to $5bn a year [26]. Whether Nvidia's cut of that is a rounding error or a co-ownership stake depends entirely on a number nobody outside the deal has seen [27].
The control language matters more than the headline pause. Nvidia told some cloud providers they could rent its chips only to approved customers, and signalled it preferred capacity spread among several smaller firms rather than parked with one large one, which irked some potential partners [7]. Nvidia's spokesperson frames that as opening compute access to a fast-growing ecosystem, a model that "continues to evolve due to high demand" [8], and it also means the supplier sits in on your sales meetings.
The programme was 57 days old when the Journal reported the step-back on 27 August [24][2]. It sits inside a much larger apparatus: $6.3bn of CoreWeave capacity backstopped in September 2025 [10], a $500bn financing platform assembled with six large financial institutions [11], a guarantee of up to $105bn to help OpenAI lease a data centre [12], and more than $40bn in AI equity stakes taken this year [13]. Investor scrutiny of those circular arrangements has been building [23], though no regulator has opened an inquiry and the pause, on the reporting, came from inside the company [20].
Anyone holding a plan that assumed this credit line is facing a forcing function now, and it comes down to two questions: does the purchase still finance without a guaranteed buyer, and is your paper already signed. Signed and financeable, you are fine and should expect a repricing conversation. Signed and not financeable, the practical step is written confirmation that your deal is not among the paused ones, since the reporting named none of them and neither Sharon AI nor Firmus has said whether its own agreement is affected [18]. Unsigned and financeable, buy on ordinary commercial terms, which is what every deep-balance-sheet buyer has done all year [21]. Unsigned and not financeable, treat the capacity as a forecast until Nvidia says which version of the programme comes back [9]. All of this rests on unnamed sources that TNW did not independently verify [14].
What to watch
- Whether Nvidia names the paused deals, or Sharon AI and Firmus confirm the status of their 40,000 and 170,000 GB300 commitments.
- Whether a revamped or folded-in version keeps the approved-customer condition that partners objected to.
- Whether any competition regulator opens an inquiry, given that none had as of the reporting.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence50
- Adoption30
- Hype gap+15
- Incentives70
- Confidence55
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Nvidia has halted some of the transactions under the financing programme it introduced eight weeks earlier, under which it extends credit to AI cloud companies buying its chips and takes a cut of what those chips later earn; the programme was announced on 1 July.
- [2]
The Wall Street Journal reported on 27 August, citing people familiar with the matter, that Nvidia had stepped back from the programme the previous week.
- [3]
Under the arrangement Nvidia sells the chips, guarantees to rent back capacity that a customer cannot resell, and then collects a share of the cloud revenue those chips generate, so it is paid at the point of sale and again along the way.
- [4]
Nvidia sought to rent compute capacity back from cloud customers if they were unable to sell it, providing them with a guaranteed buyer and making it easier for the firms to borrow the capital needed to fund purchases of Nvidia's AI chips.
- [5]
Two objections surfaced inside Nvidia: employees flagged that the structure could attract antitrust scrutiny, and raised the related question of how far Nvidia could reasonably go in dictating how its customers run their businesses.
- [6]
Some Nvidia employees expressed concerns to current and potential customers that the initiative could draw antitrust scrutiny.
- [7]
Nvidia told some cloud providers they could rent its chips out only to approved customers and signalled it preferred the capacity be distributed among multiple smaller firms rather than one large customer; in the early weeks the extent of control it sought irked some potential partners.
- [8]
Nvidia did not confirm the pause but did not dispute the programme's existence; a spokesperson said the new business model 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.
- [9]
The Journal reported that Nvidia could still revamp the initiative later or fold it into another programme.
- [10]
Nvidia backstopped $6.3bn of CoreWeave capacity in September 2025.
- [11]
Nvidia assembled a $500bn financing platform with six large financial institutions behind it.
- [12]
Nvidia agreed to guarantee up to $105bn to help OpenAI lease a large data centre.
- [13]
Nvidia holds more than $40bn in AI equity stakes taken this year.
- [14]
TNW has not independently verified the report, which rests on unnamed sources.
- [15]
Nvidia designed the programme to unlock purchases by cloud operators who could not raise the capital to buy chips outright, and a pause removes that route for the companies least able to find another one.
- [16]
The two deals announced at launch were Sharon AI, an Australian operator, signing for up to 40,000 Grace Blackwell GB300 chips, and Firmus taking commitments for up to 170,000.
- [17]
Firmus said at the time that it expected between $25bn and $30bn in committed customer offtake during the first six years of its deployment.
- [18]
Neither Sharon AI nor Firmus has said publicly whether its own agreement is among those affected, and the reporting did not name the paused deals.
- [19]
Sharon AI had raised $1.6bn in a private placement in June to fund its side of the deal.
- [20]
No regulator has opened an inquiry into the programme and none of the reporting suggests one is imminent; the pause came from inside the company.
- [21]
The hyperscalers are unaffected because they were never the target; anyone with a balance sheet deep enough to buy accelerators outright has been doing so on ordinary commercial terms all year.
- [22]
Nvidia said on its earnings call this week that the model had the potential to drive billions in revenue over the medium to long term.
- [23]
Investor scrutiny has mounted as Nvidia pumps money back into the AI ecosystem, stoking fears over its role in so-called circular deals that could artificially inflate demand.
- [24]
The programme was 57 days old at the date of the Journal's report.
- [25]
The two launch deals cover roughly 210,000 accelerators between them.
- [26]
Firmus's stated offtake expectation averages roughly $4.2bn to $5bn a year.
- [27]
The revenue-share percentages have never been disclosed, which is one reason outside analysis of the programme has been thin; what is known is the shape rather than the economics.
- [28]
Under the proposed deals Nvidia would receive 50 per cent of any revenue cloud providers earned through its chips beyond a certain threshold.
Get this ledger inside ChatGPT. Ask about any of our stories and the answer brings its figure, each claim’s verdict and the sources. Free, no account.
Sources
2 independent publishers whose own reporting we read for this story.
- straitstimes.comNvidia pauses revenue sharing deals with AI cloud companies: Report | The Straits Times
1 article · August 27, 2026
- thenextweb.comNvidia has paused parts of the revenue-sharing programme it launched in July
1 article · August 28, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.