AI version of financial innovation: The seller's guarantee model behind the sharp decline of Broadcom Inc. (AVGO.US)

AI infrastructure financing is undergoing a structural transformation, and the market is starting to price in this change. On August 14, Broadcom Inc. (AVGO.US) saw its stock price close down over 5.9%, with an intraday drop of nearly 7%. This sell-off was not due to a sudden decline in performance but stemmed from investors beginning to reassess the credit risks associated with Broadcom Inc.'s "seller-backed" model promoted through the AI XPV financing platform. At the same time, NVIDIA Corporation is pushing a larger-scale similar initiativereportedly collaborating with six Wall Street institutions including Apollo, Blackstone, BlackRock, and Goldman Sachs Group Inc. to mobilize over $500 billion in third-party capital through a computing financing platform. The actions of these two AI chip giants signify a new phase in the construction of AI infrastructure: as the capital expenditures of hyperscale cloud providers approach their financial limits, chip sellers are proactively engaging by providing guarantees for special purpose vehicles (SPVs) to securitize expensive computing assets and leverage funds from the private credit market. This has raised concerns in the market: is this model a structured extension of AI capital expenditures, or a new form of financial leverage risk? Hyperscale cloud providers hit capital expenditure limits, spurring innovation in financing structures According to news from the Chase trading desk, a Barclays research report indicates the direct trigger for new financing demands in AI infrastructure is that hyperscale cloud providers are approaching their natural limits on capital expenditures. Barclays estimates that the combined capital expenditures of Amazon.com Inc., Alphabet Inc. Class C, Meta, Microsoft Corporation, and Oracle Corporation will exceed their operating cash flow by 2026, with the "funding gap" expected to widen to approximately $210 billion by 2027, and further increasing in 2028. The report points out that hyperscale cloud providers are facing real constraints in terms of debt levels and power agreements, with bond issuance having significantly expanded, and some companies seeing their capital expenditures exceed 100% of their operating cash flow. Meanwhile, AI labs are experiencing extremely strong revenue growth. Barclays estimates that the annual recurring revenue (ARR) for AI labs is expected to exceed $200 billion by the end of 2026, roughly twice and five times the previous forecast values by OpenAI and Anthropic, respectively. The tension between the strong demand for computing power and the tightening capital expenditures of cloud providers has created conditions for the emergence of new financing structures. In this context, a structure that separates the "shell" of data centers from the financing of computing assets through SPVs is beginning to take shape. In 2026, the cost of constructing a 1GW data center is estimated to include approximately $15 billion in non-computing assets and about $35 billion in computing assets. The latter, being larger in amount, depreciating faster, and carrying higher obsolescence risks, is also the core issue that the new securitization structure aims to address. Broadcom Inc. XPV platform: Providing a "backstop" for private credit as a chip seller Broadcom Inc.'s AI XPV Platform was officially launched in June this year, co-established by Broadcom Inc., Apollo, and Blackstone, with an initial capital scheme of $35 billion, aiming to support over 20GW of AI computing capacity by 2028. According to the Barclays research report, the basic logic of this financing structure is as follows: AI labs such as Anthropic or emerging cloud service providers (neocloud) purchase TPU computing units co-designed by Broadcom Inc. and commissioned by Alphabet Inc. Class C, manufactured by Taiwan Semiconductor Manufacturing Co., Ltd. (TSMC), with the related computing assets injected into the SPV; Broadcom Inc. provides approximately 85% guarantees for the priority notes issued by the SPV, while Blackstone or other financial institutions then provide investment-grade financing for this SPV; Anthropic subsequently pays the SPV rental fees for the computing power, with cluster operation management handled by neocloud such as Fluidstack. This structure offers clear strategic value for Broadcom Inc.: customers can quickly deploy computing power without incurring high upfront capital expenditures, while Broadcom Inc. can leverage the funds from financial institutions to amplify the market penetration of its XPU, further expanding AI chip revenue. However, Broadcom Inc.'s role in providing guarantees for debts as a seller also means that its balance sheet will bear potential or contingent liabilities as the platform scales up. Barclays estimates that Broadcom Inc.'s cumulative guarantee exposure could approach $739 billion by 2028. NVIDIA Corporation's $500 billion plan: Similar logic, larger scale As Broadcom Inc.'s XPV platform draws market attention, NVIDIA Corporation also announced this week a larger-scale initiative of a similar nature. According to an announcement from NVIDIA Corporation, it has reached cooperation intentions with Apollo, Blackstone, BlackRock, Goldman Sachs Group Inc., Brookfield, and KKR to jointly establish an AI computing infrastructure financing platform aiming to mobilize over $500 billion in third-party capital, corresponding to approximately 8 to 10GW of computing deployment. According to the Barclays research report, the basic structure of NVIDIA Corporation's platform is similar to that of Broadcom Inc.: neocloud purchases GPUs and injects them into the SPV, with NVIDIA Corporation providing guarantees of no more than 25% for the SPV's priority debt, thereby reducing financing costs; end-users like OpenAI rent GPU computing power from neocloud. Unlike Broadcom Inc., NVIDIA Corporation not only provides guarantees but also retains the rights to share rental income exceeding the preset hourly fee. Barclays expects the new AI securitization market to expand rapidly: by 2027, it could account for about 20% of overall industry capital expenditures, and if progress goes smoothly, it could reach nearly 50% by 2028. Meanwhile, the related guarantee exposure is also expected to gradually reflect in the financial documents of companies such as Alphabet Inc. Class C, NVIDIA Corporation, and Broadcom Inc.the procurement obligations disclosed by Alphabet Inc. Class C in its Q2 2026 10-Q document have reached $811 billion, and Barclays believes that about half of this can be attributed to data center guarantee exposure. Market repricing: Guarantee scale, chip residual value, and lessee concentration Broadcom Inc.'s stock price drop reflects the market's reassessment of the credit risks associated with the aforementioned model. Bank of America Corp. analyst Tom Curcuruto pointed out that the current concentration of lessees on the XPV platform is quite high, with the first transaction primarily relying on Anthropic. Although OpenAI may become a future customer, other lessees have not yet been clarified. If the platform's lessees are too concentrated, should any core customer encounter repayment issues, the related SPV would face significant pressure. Moreover, the uncertainty regarding the residual value of customized AI chips also poses a potential risk. Unlike assets such as aircraft or servers that have well-established secondary markets, Broadcom Inc.'s customized XPUs lack adequate liquidity support. In the event of customer default, the pricing and speed of disposing of related chips are highly uncertain, which would directly impact the safety net of SPV financing. The Barclays research team has provided a relatively positive assessment of these concerns in its report. The report believes that there is a high degree of information transparency and an interest coordination mechanism among AI chip sellers, manufacturers, cloud providers, and AI labs, making the probability of large-scale defaults relatively limited. Furthermore, the high versatility of NVIDIA Corporation's GPUs means that even if demand weakens in one area, computing resources can still be reallocated to other scenarios where demand remains strong. However, the most pressing underlying question in the market right now is: if maintaining this wave of AI capital expenditure boom requires increasing amounts of debt financing and seller guarantees to support it, how solid is the real cash flow basis for this growth? The answer to this question may continue to influence the valuation trends of the AI chip sector. This article is reprinted from "Wall Street Insight," author: Zhang Yaqi; edited by GMTEight: Yan Wencai.