As Nvidia prepares to release its earnings, analysts anticipate its revenue growth will continue to outpace most peers. Yet a Morgan Stanley corporate credit team report on the chipmaker raises a cautionary flag: for a $5 trillion "cash machine," a customer subsidy program measured in the hundreds of billions might seem trivial, but it still warrants vigilance.
Analysts Lindsey Taylor and Nishant Satyam have initiated coverage of NVIDIA with a neutral rating—a designation widely understood within the industry for its implicit implications. The report praises Nvidia for "converting balance sheet strength into a strategic financing tool for AI," but simultaneously warns: "Tail risks are still in early stages, opaque, and large in scale—hardly a reason to turn bullish prematurely." The rationale centers on the fact that ecosystem-related support exists in the form of contingent liabilities, contractual obligations, and potentially off-balance-sheet arrangements, meaning traditional leverage metrics increasingly fail to capture Nvidia's full credit picture. The analysts have dubbed this phenomenon "balance sheet as a service."
Where to Begin
The core question the Morgan Stanley analysts seek to answer is: what proportion of Nvidia's support packages and lease commitments should actually be classified as debt? This issue came into sharp focus this month when Nvidia announced its "replicable financing platform" agreement. The company partnered with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to raise more than $500 billion for AI infrastructure construction. In a company blog post, CEO Jensen Huang stated the fundraising effort aims to address market concerns about circular financing. Bringing in a roster of top asset managers marks "the beginning of an open capital market for AI infrastructure," with Nvidia assuming only a "residual value backstop mechanism" that caps its exposure at 25% of any single project's risk.
Even a non-professional analyst can calculate the implications: 25% of $500 billion remains a colossal sum. However, since the AI infrastructure capital market currently exists largely as a collection of memorandums of understanding, estimating Nvidia's potential credit risk exposure requires a degree of speculation. The Morgan Stanley analysts use a $35 billion chip leasing agreement as a reference point. That deal, led by Morgan Stanley, involves Broadcom leasing tensor processing units—co-designed with Google—to Anthropic. The operational structure works as follows: Broadcom sells the chips to a private credit vehicle, thereby moving the associated costs off its balance sheet; that vehicle is funded by debt issued by Apollo and BlackRock, with most of the debt guaranteed by Broadcom.
Why the Focus on Just Ten Stocks
Morgan Stanley estimates that if Nvidia signs 15 similar financing agreements by the end of 2028, factoring in drawdowns and amortization, its tail risk exposure would peak at nearly $90 billion shortly after 2028. But with limited public disclosure, rating agencies tend to treat roughly $125 billion of Nvidia's various financing arrangements as debt-like. The analysts are more concerned about agreements Nvidia reaches with individual customers, partly because outsiders know very little about the specifics. In a blog post last month, Nvidia unveiled a plan to "massively unlock AI computing power," aiming to extend chip supply beyond hyperscale cloud providers. The prevailing speculation among industry observers is that, in addition to conventional product sales revenue, Nvidia will offer new AI compute cloud service providers a floor price for hourly GPU rentals; in exchange, Nvidia would share in any revenue above that guaranteed baseline.