At peak euphoria, flirtations come easy.
On September 28th, at The Korea Society’s annual gala dinner in New York, the founder, president and CEO of chip maker Nvidia had the room laughing over his Korean escapade, when he mentioned how an overexcited employee of Samsung kissed him on the lips. “I’ve only been kissed in the mouth by a man in Korea,” said Jensen Huang, amid peals of laughter. He went on to describe how he had also downed somaek [a South Korean cocktail made with beer and distilled alcohol from rice].
The fact that the anchor, too, referred to him as a 'rockstar', shows the excitement around Nvidia and Huang given that it’s the world’s most valued firm at $5.19 trillion, and unarguably the poster boy of the AI boom.
While the human kiss was accidental, Nvidia is having a serious fling with its customers.
The company’s quarterly filing for the second quarter ended July 2026 shows the company owns stakes in the businesses that buy its chips, renting back their capacity, guaranteeing rent and giving some clients longer time to pay back.
None of the above is illegal or forbidden but it’s the growing proximity and the intensity that makes the numbers a bit unsettling.
Nvidia’s support for its customers grew about five-fold in just six months of this year, from roughly $55 billion at the start of the year to roughly $288 billion, according to Fortune India estimates, based on the disclosure in the filings. [The company did not respond to questions sent by Fortune India on the contours of the arrangement.]
The $288 billion comprises equity stakes worth $98.9 billion that Nvidia already holds in companies, of which $47.7 billion are in listed companies (including $5.0 billion locked up until December 2027), $47.9 billion in private companies and $3.3 billion in infrastructure financiers. Besides, the chip maker has committed to $25 billion of equity investments, $36 billion of computing time it has promised to rent from AI cloud companies over a six-year period. These firms buy Nvidia chips, build data centres and rent the computing power out; whatever they cannot rent to other customers, Nvidia has agreed to pay for. In addition, the chip maker has $20 billion worth of data centre leases it has signed to hand on to third parties. The biggest one is the $108.5 billion guarantees, comprising $105 billion for OpenAI's Ohio campus and $3.5 billion for other AI cloud partners. Interestingly, SB Energy, the renewable energy and AI data center infra platform backed by the SoftBank Group, OpenAI, Ares Management, and Nvidia, will build and manage the Ohio facility, through a 20-year lease to OpenAI.
On the OpenAI arrangement, Nvidia mentions in the filing: “In exchange for the guarantees, the site will exclusively Nvidia AI infrastructure, subject to limited exceptions.” The guarantees end “upon certain events, including OpenAI achieving a satisfactory credit rating.” Currently, Open AI does not have a credit rating and in October 2026, AIR Platforms, an AI-driven rating firm, rated OpenAI BB+ speculative grade (junk), citing negative EBITDA margins. Simply put, Nvidia is lending its customers its own credit standing until a customer earns one.
Why is Nvidia doing this
The filing explains the rationale directly: AI clouds and model makers currently lack the ability to secure long-term infrastructures contracts and investment grade financing capacity to secure the AI infrastructure necessary to grow. It also means that lenders are unwilling to fund the customers yet on their own.
Interesting, in the 2026 annual report mentioned that Nvidia has been “asked for financing arrangements” and that “We have not entered into any financing arrangements.”
Six months later, the July 2026 filing states: “Financing arrangements with certain investment-grade customers, including extended payment terms under large, multi-quarter agreements, will continue to affect the timing of our operating cash flows.” Receivables rose to $24.6 billion in the first half while operating cash flow was $74.4 billion, against net income of $118 billion.
While the filing states: “We estimated that one AI research and deployment company contributed a meaningful amount of our revenue by purchasing cloud services from our customers,” it doesn’t name the company though its description fits OpenAI, the same company behind the $105 billion guarantee.
What’s pertinent to note is that equity stakes Nvidia hold is also beefing up its income statement. Unrealised gains of $23.7 billion from equity securities were added in the first half of the current fiscal ending January 2027, and about 17% of pre-tax income, against $2.1 billion a year earlier.
During the quarter, Nvidia also began a new business model with select AI cloud partners. The filing states: “AI clouds procure our data centre infrastructure products, and we commit to cloud service agreements, which the AI clouds can unilaterally stop providing to us and sell to third-party customers at more advantageous rates.”
“If you look at the fundamentals, the demand for the neo clouds is off the charts…what is missing in the middle is the working capital to fuel that growth,” Huang told CNBC in an interview.
The risk factors in filing spell out who carries the downside: “If AI clouds do not successfully sell committed capacity to third-party customers, we have agreed to purchase that capacity.” In other words, Nvidia carries the cost of weak demand.
The CoreWeave stress test
What makes Nvidia strategy tricky is the current travails of CoreWeave, which provides high-performance AI computing infrastructure. Nvidia is its shareholder, its only GPU supplier, its customer and the buyer of last resort for capacity it can’t sell.
CoreWeave’s 2025 annual report says that “all of the GPUs used in our infrastructure today are NVIDIA GPUs.” Nvidia also appears among its customers, under a master services agreement dated April 2023. In September 2025, Nvidia placed an order with an initial value of $6.3 billion. CoreWeave's filings say Nvidia “is obligated to purchase the residual unsold capacity through April 13, 2032.” In January 2026, Nvidia bought another $2 billion of CoreWeave shares at $87.20 each, taking its holding to about 47 million shares, roughly 11% of the company.
The month the market tested the valuation: CoreWeave shares fell 15.5% on October 14 to about $66, as yields on its unsecured bonds rose toward 15–16%. The company’s debt had already grown from $8 billion to $21.6 billion, and $50 billion debt as of mid-2026.
At $62 a share, Nvidia’s 11.50% stake in CoreWeave is roughly break-even on cost: its 47.2 million shares are worth about $2.93 billion, against roughly $2.97 billion paid. Early shares bought at about $40 or less show a gain of some $530 million. The 22.9 million bought in January at $87.20 are down about $580 million, or nearly 29%. The fall is sharper since June. Nvidia’s 13F filing shows 47.2 million shares on June 30, when CoreWeave traded near $99.60, putting the stake at about $4.7 billion. At $62, that is a write-down of close to $1.8 billion that will flow through the P&L, if the price holds through the quarter. It may appear a small change given Nvidia’s quarterly profit of $59.7 billion. But what is not visible now is the pain that could build up.
If CoreWeave’s finds demand weakening, Nvidia is obligated to buy unsold capacity. Microsoft is the biggest customer, making up for 67% of CoreWeave's 2025 revenue. Microsoft’s share though has fallen with Meta, OpenAI and others signing newer large deals. While CoreWeave’s largest customer is now 36% of revenue (Q2 2027), if long contracts that come up for renewal don’t play out as expected, Nvidia’s backstop will get triggered. [Backstop is like a minimum revenue guarantee that entails Nvidia absorbing part of the shortfall if a customer’s cloud capacity is not fully rented out]
The case for the underwriting
The strongest defence of Huang's strategy is the size of the business. In the first of the current fiscal, Nvidia clocked revenue of nearly $173 billion, operating income of $63.7 billion and net income of $118 billion.
Much of the exposure is also deferred and capped. The backstops are not due in the remainder of the current fiscal. The OpenAI guarantees phase in from fiscal 2029, and the filing says they “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.” OpenAI “has agreed to reimburse and indemnify us for certain losses,” though the filing adds that “we may not recover amounts promptly or in full.”
Nvidia's revenue is also less concentrated than a year ago. Its largest direct customer was 16% of quarterly revenue, down from 23%. The extended payment terms, by the filing's own description, are for “investment-grade customers,” not start-ups. And Nvidia says it will apply discipline: it intends to “focus our data center infrastructure initiatives on exceptional sites and apply the same discipline to these initiatives as we do to our supply-chain management.”
Nvidia knows that it cannot play the game alone and in August 2026 had an understanding with large capital providers to establish independent financing platform to garner over $500 billion of third-party capital over time to support the deployment of AI infrastructure. But there is no sign of such a pool of capital coming up. “These and other preliminary arrangements may not lead to definitive agreements,” Nvidia stated in its filings.
Is Nvidia entering a forbidden circle?
Given the scale of its commitments, Nvidia now owns, backs, rents from and extends credit to its chip customers.
For instance, media reports mention that AWS is taking deliveries of $5 billion worth of Nvidia GB300 server racks at its U.S. facilities, with an additional $3 billion of hardware to be purchased each month in the foreseeable future. The issue is lack of grid connectivity has resulted in idle servers. The aggressive purchase has for the first time resulted in negative free cash flow for the company.
With leverage funding the demand for chips across big tech and rising interest rates make the equation dicey. If AI demand slows, Nvidia could lose orders, mark down its stakes, buy unsold capacity, extend more time to pay and face guarantee calls, all in the same quarters.
For a stock that has surged 10x in the past five years to hit all-time highs, valuations are not exactly in bubble territory at 27 times trailing 12-month earnings. According to analyst estimates, Nvidia's revenue is set to compound 61% over FY26-FY29, with earnings outpacing at 65% over the same period. That’s enough to keep the momentum going, the hitch is that a lot of it is baked into future earnings.
While the party lasts, Nvidia’s circle of influence on customers will continue to expand given that last fiscal it generated $97 billion. But of every $100 of free cash flow Nvidia generated, about $33 went into customers and partners, $43 went back to shareholders, and what was left was just $24. Not surprising that Morgan Stanley has a neutral credit rating on Nvidia, stating: “…exceptional growth turns balance-sheet strength into a strategic AI financing tool and introduces new risks,” adding that the tail “remains too early-stage, opaque, and sizable to step in.”
The bigger risk Morgan Stanley sees for Nvidia is that’s total credit exposure will hit nearly $200 billion by the end of 2028, with $170 billion in guarantees and backstops that are off the balance sheet today but will hurt in the event of a downturn.
Investors are drawing parallels of such circular financing with the telecom bubble at the turn of the millennium when telcos financed the build-out with roughly $1 trillion debt for laying fibre cables, equipment, M&As. When the 2000 dotcom bubble burst, reports indicate 80 million miles of fibre was ready but unused. The subsequent Nasdaq meltdown coincided with the implosion of WorldCom.
Investors are questioning whether such dependencies could pose risks if AI demand or monetisation falls short of high investor expectations.
At the gala dinner, Huang did mention that “nobody suffers better than I can,” in the context of Nvidia’s near bankruptcy in the late ’90s. While the Nvidia of today is no longer the fragile start-up of yesteryear, history does rhyme. At some point, the AI trance music will stop, and Nvidia's filings show why the fling could prove costly.