⚡ Key Takeaways
Surge in Credit Sales and Debt Guarantees:
Nvidia's accounts receivable soared to $63.1 billion and customer debt guarantees reached $109 billion, deepening market suspicions over "round-tripping finance," where the company directly provides funding to sustain and expand the AI ecosystem.
Wall Street-Linked Infrastructure Financing and Loss Guarantee Risks:
With Nvidia undertaking up to 25% ($125 billion) in first-loss guarantees for a $500 billion AI infrastructure fund envisioned by Wall Street financial firms, concerns are mounting that overinvestment in data centers could spill over into financial risks.
Profitability Uncertainties and AI Cycle Volatility:
If data center demand slows due to enterprise preference for cheaper AI models, the rise of low-cost Chinese AI, and persistent high interest rates, the growth momentum across the entire AI ecosystem—extending beyond Nvidia to Korean chipmakers—risks being destabilized.
* This article is based on a video published on August 28, 2026.
How long will the miraculous semiconductor boom triggered by the AI development race last? Nvidia's quarterly earnings report released early Thursday morning (KST) temporarily eased these concerns. However, the deeper one looked into the earnings sheet, the more numbers appeared that failed to fully clear away skeptical perspectives. It was still an unbelievable, staggering performance: quarterly revenue alone reached a massive $96.2 billion. That means the company is pulling in over 1.4 trillion won every single day in Korean currency.
Nvidia has beaten market expectations without fail for 3 years and 9 consecutive months. Yet this is precisely the biggest problem facing semiconductor stock prices today: expectations are already so high that exceeding them by enough to prevent disappointment has become extraordinarily difficult. While Nvidia initially blew past market expectations by forecasting that next year's revenue could jump another 70%—generating around 2.6 trillion won per day—the market has not completely shaken off its doubts, particularly because of two subsequent figures. What are these two numbers?
Colette Kress | CFO of Nvidia
We are fully aware of the scale of financial support (that Nvidia is providing to customers). We know some call it 'round-tripping finance.' But we see it differently.
* Source: NVIDIA
At the end of the day, those who doubt the sustainability of AI development are asking this: "Who on earth has the money to keep buying those expensive AI semiconductors in such massive quantities? Can hyperscalers like Google and Amazon carry this alone?" The answer is no. Looking at Nvidia's latest earnings release, U.S. hyperscalers such as Google and Amazon do account for 55% of Nvidia's data center revenue. These giants appear unlikely to slow their pace in the AI race even if it eats into their own margins, and for the most part, they have the financial muscle to sustain it.
However, the remaining 45%—customers outside the hyperscalers—must also be able to stably and consistently spend vast sums on semiconductors without stumbling for Nvidia's earnings, the current chip boom, and the performance of Samsung Electronics and SK Hynix to keep growing. Can they pull that off? Early Thursday morning, Nvidia answered with a firm "yes." In the market, however, debates over this issue began attaching themselves to two specific numbers that followed the stellar headline earnings.
Nvidia's revenue for the quarter more than doubled from a year earlier. In contrast, Nvidia's free cash flow stood at $21.3 billion, an increase of over 58.7% year-over-year. While that remains an impressive figure and provides relief compared to other Big Tech firms burning through cash on AI investments, the pace of free cash flow growth was visibly smaller than its revenue expansion. More importantly, compared to the immediately preceding first quarter, free cash flow plunged by more than half.
Why? Unpacking the financial statements makes the reason clearer. While revenue grew substantially, a major factor was that products Nvidia delivered to customers on credit surged. The value of AI accelerators shipped without payment yet collected exceeded $63.1 billion, or more than 88 trillion won. This represents a sharp increase from the 54 trillion won level recorded in the previous quarter. In response, Nvidia explained that it decided to grant extended payment terms to prime customers. Consequently, the collection period for accounts receivable stretched from approximately 45 days in the previous quarter to 60 days in the second quarter. This is the first number that made investors tilt their heads in doubt.
Nvidia disclosed another detail: for prime customers making substantial capital outlays on data centers, payment terms could be extended anywhere from 90 days up to a maximum of one year. In essence, it can offer up to a year of credit to customers. Of course, this alone cannot be interpreted as a sign that the roaring semiconductor boom is drawing to a close. While Nvidia did not specify which firms hold these credits, it left hints in the earnings release that they are hyperscalers. The company's message is that the money will arrive soon and there is nothing to worry about.
Nevertheless, it is clearly evident that the number of players across the AI investment landscape requiring Nvidia's patience is rapidly increasing. The perception that "Nvidia cannot merely sell its products, but must also grant extensive credit terms in certain cases" has reignited lingering market concerns over the financial and credit assistance Nvidia has disbursed across its customer base—the so-called round-tripping finance.
AI Ecosystem's 'Financier': Passing the Torch from Nvidia to Wall Street?
It is no exaggeration to say that artificial intelligence worldwide runs on Nvidia's AI accelerators. Yet lately, there is something Nvidia is pursuing just as passionately as developing these accelerators: becoming the primary financier and bank of the AI ecosystem. The implicit message is: "I will provide the funds, so let the AI ecosystem continue to expand rapidly. Build scale, and make sure that scale is built using our Nvidia AI accelerators." Although Nvidia currently earns over 1.4 trillion won a day, this is already baked into its stock price. For the current valuation to be justified, the market needs confidence that next year the company will bring in more than 2.6 trillion won daily.
This is why market discussions about AI circular financing have persisted since last year: "I will provide you with funds, and you use that money to purchase my products." First, Nvidia has invested in companies that purchase its AI accelerators, effectively financing them to buy more chips and construct more data centers. Second, if customers worry that nobody will rent the data centers they built using Nvidia chips, Nvidia steps in as a customer: "I will utilize the surplus cloud capacity and computing power in the data centers you build." The agreement between Nvidia and CoreWeave, a prominent firm specializing in leasing Nvidia AI accelerators, represents precisely this arrangement.
This brings us to the second number the market focused on in this earnings release. Barron's analyzed that the volume of debt guarantees Nvidia indicated it could assume for customer obligations reached $109 billion this quarter. The vast majority of these guarantees were extended to OpenAI. In the event OpenAI leases and fails to adequately meet rental payments for a data center currently breaking ground in Ohio, Nvidia agreed to cover up to $105 billion. Witnessing this dynamic, questions began surfacing across the market: "Isn't this just passing money back and forth among yourselves? Is there any genuine, newly created profit here?"
Jensen Huang's answer to this had been announced earlier: Wall Street will finance the growth of the AI ecosystem. A staggering $500 billion—exceeding 700 trillion won in Korean currency—could be mobilized from Wall Street rather than coming directly from Nvidia's coffers. At 700 trillion won, this exceeds the total capital expenditures made last year by America's top six tech giants, the six hyperscalers like Google and Amazon driving AI investment demand. Does this mean genuine new capital is finally entering the AI ecosystem?
"A Fantastic New Financial Instrument" Backed by Nvidia Products?
Larry Fink | BlackRock CEO (Local Time August 10, 2026)
(The 'AI infrastructure investment platform') can be a fantastic investment product. It will broaden high-quality, long-dated earning investment opportunities.
* Source: CNBC
These are the words of BlackRock CEO Larry Fink, often dubbed the "true emperor of Wall Street." Six heavyweight financial institutions, including Larry Fink's BlackRock—the world's largest asset manager—along with Goldman Sachs and Blackstone, are the Wall Street giants looking to bring more than $500 billion into AI infrastructure investments, just as Jensen Huang noted. How does this connect to individual investor capital, and how would it operate?
While the exact financial structure has not been fully finalized, investment products backed by Nvidia AI accelerators and the future usage fees generated by running them appear imminent. The most prominent scenario currently under discussion is as follows: six financial firms, including BlackRock, establish special purpose vehicles (SPVs), to which Nvidia sells AI accelerators. These SPVs then lease the acquired AI accelerators to data centers that may find purchasing expensive accelerators outright burdensome. In exchange, the Wall Street-created SPVs collect lease payments. Wall Street would then package these into asset-backed securities and sell them to investors. This would allow investors to participate in securities offering steady cash flow from regular monthly usage fees. The prevailing scenario mirrors mortgage-backed securities (MBS), a market Larry Fink helped pioneer in the 1970s, now adapted to data centers to spur massive investment.
Larry Fink | BlackRock CEO (Local Time August 10, 2026)
It's very early days, like when I started working in the mortgage-backed securities market in the '70s. I think this could be the next generation of opportunities in financial product development.
* Source: CNBC
Escaping 'Round-Tripping Finance' vs. Transferring 'AI Risks' to Financial Markets
What happens, however, if these data centers are not fully utilized? These data centers are leased by AI developers. While hyperscalers build their own facilities, companies like OpenAI and Anthropic also lease data center capacity. Beyond OpenAI and Anthropic, firms worldwide eager to enter the AI race typically start with leased data centers. What if, after spending enormous capital to purchase or lease AI accelerators and construct facilities, an insufficient customer base emerges? What if data centers subsequently fall behind on lease payments? If newly erected data centers fail to generate revenue, what happens to the securities Wall Street sold with those data centers as collateral?
While mortgage-backed securities (MBS) are backed by residential properties, these securities would be collateralized by AI infrastructure, including AI accelerators. Yet electronic hardware rapidly depreciates after several years as newer, more advanced models continuously hit the market. Although Nvidia's AI accelerators differ from everyday consumer electronics, Nvidia also releases new products on a regular schedule, inevitably depreciating older hardware over time. In fact, Thursday morning's earnings report showed that Nvidia's inventory management costs rose significantly compared to the prior quarter following the introduction of its latest AI accelerator.
"Even for Nvidia AI accelerators, inventory builds up like this when new products launch. How long, then, is the lifespan of an Nvidia AI accelerator, and how long can it maintain its value?" Opinions on this matter diverge sharply. Jensen Huang argues that AI accelerators will easily last 10 years. In contrast, Wall Street investor Michael Burry, who holds a more pessimistic view of the semiconductor sector, estimates their viability at just 3 years. Wall Street consensus generally hovers around 5 years. Many enterprise loans in the AI sector carry five-year terms, meaning that if collateral value holds steady for 5 years, major disruptions could be averted.
What if that does not happen? What if investors who trusted Nvidia and Wall Street to purchase these new securities incur losses? This is where Nvidia's reported agreement with Wall Street enters the picture: Nvidia may guarantee up to 25% of the $500 billion Wall Street intends to raise. Nvidia could guarantee the residual value of the collateralized AI accelerators and infrastructure up to roughly $125 billion. In essence, while tapping Wall Street capital, Nvidia is stepping in to partially back the secondary market value of its products if necessary. Nvidia cannot simply put up an additional $500 billion on its own, but its pitch to Wall Street and institutional investors—such as pension funds and insurance companies—effectively says: "Nvidia will provide a first-loss guarantee up to about $125 billion, so please participate in this investment."
"Do We Really Need Overly Expensive AI Models?"
If this generates stable cash flow through sound lending and healthy securities, Nvidia will not need to fulfill loss guarantees and can simply enjoy the fruits of sustained growth, as can equity investors. But what if data center capacity ends up in oversupply?
Recent trends show sluggish sales for Anthropic's most expensive model, Fable 5. Reports also suggest sales of OpenAI's highest-priced models have cooled. Enterprises are increasingly concluding: "Cheaper models than Fable 5 handle our tasks just fine. Top-tier models are simply too expensive." Furthermore, while still a distant prospect, if low-cost AI from regions like China demonstrates capable performance and spreads despite security concerns, the frantic pace of the current AI development race could decelerate. Under such conditions, sentiment could quickly shift toward viewing current data center construction rates as excessive.
Even if American hyperscalers continue aggressively competing to become ultimate winners, non-hyperscaler entities that account for the rest of the global AI infrastructure race—and are essential for Nvidia to sustain its present growth velocity—could lose momentum much faster.
A persistent or escalating high interest rate environment poses another challenge. Many AI-related firms face steadily rising capital costs to finance investments in data centers and infrastructure. In this climate, investors holding securities collateralized by AI accelerators and infrastructure might complain: "Other assets offer higher returns now; is this all the yield we get?" While it appears unlikely at present that Wall Street investment products with Nvidia guaranteeing up to 25% of losses would trigger systemic financial instability, even a partial realization of these concerns could undermine the rapid rally of Nvidia and the chip market, leaving current stock valuations hard to justify. If that occurs, the AI accelerator-backed securities Larry Fink touted as fantastic investments will struggle to be perceived as the safe, assured instruments promised today.
Park Sang-hyun | Senior Specialist at iM Securities
The positive takeaway (from Nvidia's latest earnings) is that it eased concerns regarding slowing demand. It also provided some relief regarding worries over round-tripping finance. (However, going forward,) if operational performance stumbles even slightly, the possibility remains that debt repayment pressures could mount significantly. Therefore, if GPU pricing decelerates or an unexpectedly severe high-interest-rate scenario unfolds, all risks could concentrate on Nvidia and potentially shake the entire AI cycle, which warrants sufficient vigilance on our part.
※
Copying, redistribution, and unauthorized use in AI training are strictly prohibited.