Nvidia (NVDA.US) $500 billion AI finance bureau restarts: Big Six Wall Street partners and zero contracts officially announced, guaranteeing up to 25% of the residual value of their chips

The Zhitong Finance App learned that in mid-August, Goldman Sachs, Blackstone, and Apollo purchased a Nvidia (NVDA.US) chip design debt financing plan for AI developers, which had been operating secretly for several months. At a time when progress was slow, Nvidia CEO Huang Renxun chose to disclose this effort: a total of six institutions aimed at financing a $500 billion AI computing power transaction — an integer target with unclear origins. After a month, the deal became more clear.

Months of conspiracy and “add people” a few days ago

A behind-the-scenes review showed that behind this momentum was a hasty alliance: Goldman Sachs, Blackstone, and Apollo's debt plans were slow to come to fruition, so Wong In-hoon changed the strategy and announced it first. The move is intended to appease Nvidia investors — a number of well-funded institutions are ready to finance their customers, particularly key AI startups such as Anthropic and OpenAI that are relevant to Nvidia's future needs. Hwang In-hoon is optimistic about AI spending as a whole, but Nvidia has been seeking to expand its customer base beyond hyperscale cloud vendors such as Microsoft and Amazon — many of which are developing their own chips.

Hwang In-hoon wanted something else. After months of cooperating with three financial institutions, the $5.5 trillion company called the original team a few days before the announcement to inform them that KKR, BlackRock, and Brookfield would also join and promise to finance part of the debt.

After the cooperation is revealed, Wall Street's largest institutions will arrange hundreds of billions of dollars of financing for chip transactions at any time, and Nvidia itself will provide guarantees and endorsements for some of these transactions. However, according to people familiar with the matter, not a single deal was signed at the time the announcement was made, and the announcement was deliberately vague.

The bond market was shocked and then decided: the negative line on August 10

Investors have been worried that Nvidia and others will blow a bubble into the industry through “revolving finance” — Nvidia has invested in its own customers such as CoreWeave, and such deals have heightened concerns. The initial statement about this financing arrangement made debt investors uneasy, and they feared that it would leave chipmakers burdened with higher leverage exposure; concerns abated as Hwang In-hoon clarified that Nvidia's support cap was 25% of a single opportunity and assessed on a case-by-case basis.

The first reaction of the market was “shock” rather than surprise: on the day of the announcement on August 10, Nvidia's stock price fell more than 3% intraday, closed down 2.86%, and stabilized the next day; the 5-year CDS, which measures its credit risk, once rose to 77.2 basis points, the biggest one-day increase in two weeks. Since late May, the cost of default insurance for Nvidia's debts has nearly doubled.

Wong In-hoon immediately publicly denied that the arrangement constituted revolving financing: “This move is aimed at dispelling market concerns, and we are introducing independent long-term institutional capital into the AI infrastructure market.” He emphasized that the demand is real and that the capital side will independently insure each project. Based on this, some media have estimated that if the $500 billion funding target is fully achieved, Nvidia may need to use up to 125 billion US dollars of its own capital to cover it.

“Debt Showcase”: Creating a trend with zero contracts and no timeline

Since then, Huang Renxun and six agency executives have been on the CNBC TV program for more than 30 minutes without much new details: Goldman Sachs CEO Su Dewei, Blackstone President Jonathan Gray, Apollo President Jim Zelt, and Brookfield CEO Bruce Flatt appeared in the studio; Waldemar Szlezak, KKR's global head of digital infrastructure, was present; BlackRock CEO Lawrence Fink was connected via video due to travel. In an interview, Hwang In-hoon said, “In the AI field, computing power is revenue.” This is the first time that technology chips have become an investable asset category.

These executives then turned to clients such as sovereign wealth funds, pension and insurance institutions to map out their intentions to buy related debts; they revealed in TV discussions that some of the funds may have come from retail investors. There is no set timeline for this $500 billion commitment. It is a combination of already discussed deals and forecasts of upcoming demand. Each agency can review customer credit on its own before committing.

Most of the financing will come from the private equity market, but the scale will necessarily use the open market — it is expected to be in the form of bonds issued by special purpose agencies (in a single transaction or tens of billions of dollars), which will rent chips to Nvidia customers. A participant in the announcement described Hwang In-hoon's intention as setting up a “debt window” to advertise to customers and concerned investors; if the deal fails to land or goes wrong as stated in the announcement, the reputation of the financing partner and Nvidia will be at risk.

The gamble with a residual value guarantee: the MBS analogy and the depreciation dispute

More specific: Nvidia guarantees part of the residual value of the chips installed in these projects, up to 25% of a single transaction, assessed on a project-by-project basis; if the resale or reuse value of the hardware falls short of expectations at the end of the financing period, Nvidia fills part of the gap. Huang Renxun said that this ratio is “significantly lower” than other computing power financing arrangements, and that credit assessments of customer credit, demand, utilization rate, and residual value are still completed by the capital side.

Fink's analogy is boldest: it's like “the beginning of the 1970s mortgage-backed securities market” and “the next future of financial engineering.” This metaphor is a double-edged sword — MBS later became the epicenter of the 2008 financial crisis because the underlying collateral was far less worth the hypothetical price. The Verge also pulled out old accounts: Hwang In-hoon said last year that “no one wanted Hopper” after Blackwell was mass-produced; however, the $100 billion progressive investment announced by Nvidia and OpenAI in September 2025 is still an unimplemented memorandum of understanding — MOU is not equal to landing. Motley Fool's analysis points out the crux: GPUs are updated every two years. If lenders lend for a seven-year period and the collateral becomes obsolete for three years, this economic model may quickly collapse.

Hwang In-hoon's response was in favor of the sky. He said that the A100 released in 2020 has been in business for six years now, and the economic life span is ten years, and CUDA continues to increase the output of installed hardware; as market evidence, the annual contract rent for the H100 rose from $1.70 per GPU hour in October 2025 to $2.35 in March 2026, and the price of B200 computing power is in the range of 5.30-7.05 US dollars. This is in direct contrast to Michael Burry's depreciation warning — the latter accuses hyperscale vendors of underestimating 2026-2028 depreciation by about $176 billion, and claims that a GPU upgrade cycle of two to three years cannot support a five to seven year depreciation period. On Monday, Burry pointed another finger at the $573 billion AI financing network held by the insurance industry.

Nigel Greene, CEO of DeVere Group, asked the most relentless question: “If the market truly believes that demand for Nvidia chips is strong and enduring, why does the company need to personally guarantee the resale value of its hardware to reassure lenders? Confident sellers usually don't need to back up financing their customers.” He warned that structured financing, which relies on a single party to guarantee the future value of assets, has historically included some of the most destructive credit incidents in modern financial history.

Comparative sample and those in line: Broadcom has 35 billion dollars in hand, followed by Damos with 1.5 trillion

This high-profile campaign contrasts with similar announcements from Broadcom a few weeks ago. Broadcom designated Apollo and Blackstone as anchor investors to finance plans to build more than 20 gigawatts of computing power for cutting-edge laboratories such as Anthropic and OpenAI by 2028, with potential demand of hundreds of billions of dollars; however, Broadcom already had $35 billion in financing at the time of the official announcement — Broadcom covered most of the debt for the first transaction to attract investors, and Apollo placed loans outside of Broadcom's balance sheet through structural design. Bloomberg previously reported that Blackstone has tested investors for another deal of more than $30 billion.

For financing partners, this business means expenses: Goldman Sachs is the only institution with a dedicated investment banking department out of the six, and Apollo is also expected to unlock more fees after expanding its trading business. For Goldman Sachs, this is the culmination of years of groundbreaking — Jung Min, who was promoted to co-head of its technology, media and telecommunications business last year, was responsible for covering Nvidia in San Francisco for many years, and former Goldman Sachs Nvidia analyst Toshiya Hari joined Nvidia's Investor Relations department last year.

Despite the many questions, the number of people in line is growing: J.P. Morgan Chase is discussing how to get involved (people familiar with the matter); minutes after the announcement, Morgan Stanley announced the launch of a framework to facilitate $1.5 trillion in US innovation and national security funding. AI and advanced computing topped the list. Beyond Wall Street, the other end of this funding chain is already taking shape: Ares Management counts 26 AI-related financings disclosed in the past 12 months totaling $573 billion, a significant portion of which will eventually be held by the insurance industry. Apollo President Zelter estimated the total investment demand for global AI infrastructure at more than 8 trillion US dollars. Morgan Stanley expects hyperscale manufacturers to spend 3.5 trillion US dollars from 2026-2028. The Bank of England's July “Financial Stability Report” warned that this pace is unprecedented. Impacting highly leveraged AI companies may be transmitted to the global financing environment and trigger a credit crunch.

Highline Asset Management analyst Ben Emmons's comment is probably the best footnote to the whole game: “This deal didn't eliminate the shortage of computing power — it financialized it.”