On August 14, Broadcom’s stock plummeted nearly 6%, erasing approximately $118 billion in market value. The trigger was a Bank of America report stating that if XPV, the AI chip financing platform co-founded by Broadcom, Blackstone, and Apollo, expanded to a 20-gigawatt scale, Broadcom’s guarantee exposure could reach $370 billion. The market mistakenly interpreted this guarantee cap as a debt risk, but even under extreme default scenarios, Broadcom’s actual losses would amount to only about $42 billion. The underlying issue is that rising financing costs are eroding profit margins across the AI supply chain—a valuation correction, not the end of the cycle. Cloud providers were unfairly punished due to market sentiment, while asset managers like Apollo and KKR benefited from their financing activities.Article author and source: Wall Street Journal
One number, two algorithms. To trace the origin of this $370 billion, we must return to XPV, the financing platform built jointly by Broadcom, Blackstone, and Apollo. The platform operates by having institutional investors purchase Broadcom’s custom AI chips and then lease them to clients. The first transaction, valued at $35 billion, corresponds to approximately one gigawatt of computing power, with the lessee being the AI company Anthropic. Through this arrangement, the debt incurred from purchasing the chips remains on the platform’s balance sheet and is not recorded as liability on Anthropic’s books.
Broadcom's role is to provide a guarantee for the lease. If a customer stops making rental payments, Broadcom will take possession of the equipment or resell it. In its latest 10-Q filing, Broadcom disclosed that, for the first transaction, even under the extreme scenario of complete customer default and zero residual value of the equipment, its maximum exposure is $29 billion.
The $370 billion estimated by Bank of America represents the cumulative nominal limit of Broadcom's guarantees after expanding the platform to 20 gigawatts. This is the upper limit of guarantees, which is distinct from the outstanding debt.
The same report also provided loss estimates: even in the case of 100% customer default, Broadcom’s actual loss would be approximately $42 billion; at a more realistic 25% default rate, the loss would be approximately $10.5 billion. In comparison, Broadcom’s free cash flow after dividends in 2027 is projected to reach $85 billion.
A guarantee with a loss cap of $42 billion was priced by the market as if it were $370 billion in debt—this is precisely the logic behind Friday’s 6% decline.
This is not the first time this year that Broadcom has seen a disconnect between its earnings and its stock price. In July, the company raised its AI revenue guidance by more than 200%, yet its stock still closed lower that day.
The fundamental nature of collateral has prompted the market's first reaction to draw parallels with the 2008 subprime mortgage crisis. In the subprime crisis, the transmission chain involved individuals without repayment capacity obtaining loans to purchase homes, with risks securitized and dispersed throughout the global financial system. Today, GPUs are placed into special-purpose vehicles as collateral and used to secure financing based on their expected cash flows—Wall Street refers to this instrument as a "Compute Collateralized Obligation" (CCO). The cloud computing provider CoreWeave issued the first such instrument in May of this year, with a size of $3.1 billion, underlying borrowers including OpenAI and Cohere.
Structural similarities have led some market participants to directly characterize it as a replica of collateralized debt obligations (CDOs).
Michael Burry, the real-life inspiration behind "The Big Short," publicly increased his short positions on AI-related stocks this week, warning of a market bubble. On August 10, NVIDIA announced the creation of a financing platform exceeding $500 billion in collaboration with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR. The following day, its stock price fell 2.8%, erasing approximately $70 billion in market value in a single day, while its five-year credit default swap (CDS) spread has risen nearly 90% this year.
The bears' logic chain is complete. But this analogy overlooks a fundamental difference: what exactly underlies the subprime assets.
At the core of subprime loans are individuals who lack the ability to repay. In the event of default, the collateral consists of properties that depreciate in value and lack liquidity.
At the core of computing power are chips. In response to questions about "circular financing," NVIDIA CEO Jensen Huang provided the following data: the one-year rental price for H100 chips has risen from $1.70 per GPU hour in October 2025 to $2.35 per GPU hour in March 2026; even A100 chips that have been in service for six years are still operating in production environments.
An asset whose price is rising, whose lifespan is continuously extended by the CUDA software ecosystem, and which can be transferred across customers and workloads differs fundamentally from the underlying assets in the subprime mortgage crisis. NVIDIA therefore confidently offers a guarantee of up to 25% of the loan value based on residual value.
The risk is not in leverage, but in profit margins; however, this does not mean that this financing structure is without risk. The real concern is not whether a crisis will "occur," but whether funding costs will "rise."
An overlooked fact is that since the beginning of this year, the issuance volume of AI-related bonds has reached $344 billion as of early August, exceeding the total for all of 2025 by over $200 billion. However, the vast majority of these funds have been captured by investment-grade giants such as Meta, Alphabet, and Amazon. Smaller AI companies without investment-grade ratings have instead been marginalized in the bond market. The cost of CoreWeave’s most recent loan is 550 basis points above the benchmark rate, yielding over 9%.
The significance of a financing platform lies in providing an alternative financing channel for customers who cannot access low-cost funding. The demand is real, and so is the cost—borrowing costs are rising.
A deeper issue lies in the sustainability of capital expenditures. The proportion of capital expenditures to operating cash flow for hyperscale cloud providers has risen from approximately 30% in 2022 to about 60% in 2025, and according to market consensus, will reach 100% by 2026. In other words, every additional unit of computing capacity invested henceforth will need to be funded through debt or equity financing.
Meanwhile, the growth rate of capital expenditures appears to have peaked. While total spending continues to reach new highs, the "acceleration" of growth has turned negative. This is precisely the most critical element of the 2008 logic: credit structures often break at the turning point of growth rates, without waiting for growth to return to zero.
The credit market has reacted ahead of the stock market. NVIDIA’s five-year CDS has doubled since late May and rose another nearly 6 basis points on the day the company officially announced its financing platform. Bond traders have begun pricing risk for this new asset class—“computing power collateral”—while the stock market only responded on Friday. This time lag indicates that markets are treating a genuinely new variable being rigorously priced, not merely a fleeting emotional shift.
Therefore, the assessment of this event should be layered. The probability of systemic default is low, as the underlying assets generate cash flow, retain residual value, and have appreciation potential; however, rising funding costs are a current reality that will gradually erode valuations through margin compression. This is a valuation contraction, not the end of a cycle.
The dividing line lies in the default rate. A $10.5 billion loss is manageable for Broadcom if the default rate is 25%; but a $42 billion loss, resulting from a 100% default rate and the inability to dispose of the equipment, would consume nearly half of Broadcom’s free cash flow—that would be an entirely different matter.
Sliding from "valuation compression" to "breakdown" requires two conditions to occur simultaneously: a major borrower actually defaults, and the pledged chips simultaneously depreciate to the point where no buyers are willing to take them. Regarding the former, Anthropic and OpenAI are still far from investment-grade ratings and heavily reliant on refinancing. Regarding the latter, the secondary market for chips, rising lease prices, and the CUDA ecosystem’s ability to sustain demand all continue to support residual value. The probability of both conditions triggering at once is far lower than the panic implied by Friday’s 6% drop.
Who was revalued and who was unfairly punished? According to this logic, Friday’s market movement has already provided its own answer.
Broadcom was the first to be revalued. It fell nearly 6% in a single day, while the iShares Semiconductor ETF declined only 0.7% and Intel fell 2% during the same period. This indicates that the market is independently repricing Broadcom’s balance sheet, unrelated to the broader semiconductor sector’s performance.
The ones wrongly targeted are the cloud providers. On August 11, the day NVIDIA announced its financing platform, Google fell 3.84%, marking its largest single-day drop in six months; Amazon dropped 2%; Apple, Microsoft, and Broadcom each declined more than 1%. These companies have solid balance sheets, and the financing platform merely adds another source of capital—posing no real risk. Their stock declines stem largely from market sentiment, fueled by confusion between "AI funding" and "AI debt."
The true beneficiaries are asset management firms. Apollo rose 6.26% on August 11, and KKR rose 6.88%. Financing platforms are part of their business—the larger the scale, the higher the management fee income. The market clearly distinguishes, through capital flows, who bears the risk and who collects fees.
On the same day, another clue provided corroboration: Applied Materials reported record revenue and profit for its third fiscal quarter, yet its stock still fell 5% after hours. The strong earnings coupled with the declining stock price reflect the market’s reassessment of risk premiums for every company along the AI supply chain—even those with flawless financial results.
If we return the logic to the market, the observation sequence should be: On the night the news was released, first monitor Broadcom and NVIDIA’s CDS spreads and stock prices, as they are the direct vehicles of credit exposure; the next day, observe whether asset managers such as Apollo, KKR, and Blackstone can sustain their upward momentum; over the following days, watch the performance of cloud providers and secondary-tier AI chip companies to determine whether their declines are oversold or the beginning of sentiment spillover.
The deeper implication is that the market is currently pricing in a one-time write-down equating "collateral" with "debt" due to this 6% decline. However, what it truly needs to gradually absorb is a more persistent variable: when financing becomes the primary source of marginal computing power, interest costs will be embedded into the profit margins of the entire AI supply chain. This erosion will not occur overnight nor end in a single day—it will, over the coming quarters, exert a more enduring pricing force than any single news event.
The next earnings report will reveal the next steps; focus on three key signals.
First, Broadcom’s earnings report in early September. Will the company proactively clarify the guarantee framework for the XPV platform, and will rating agencies follow up with adjustments—S&P has already indicated a tendency to treat such residual value guarantees as debt. If Broadcom voluntarily increases disclosure, it suggests it is taking this credit issue seriously; if it continues to remain ambiguous, it will constitute a genuine negative factor.
Second, this is the first actual fundraising on NVIDIA’s $500 billion financing platform. A memorandum of understanding is not funding—only realized capital raises can prove this model viable. The ease of fundraising and the interest rates will directly determine whether the narrative of "computing power as collateral" can be sustained.
Third, the capital expenditure guidance from the four major cloud providers for the next quarter. If the growth turning point shifts from "peaking" to "declining," the assessment in this article would need to be revised—