Wall Street Doubts AI Spending Boom Can Pay Its Bills

The headline debt numbers investors see are not the real number. Across Meta, Alphabet, Microsoft, Oracle and Amazon, roughly $1.65 trillion in AI-related obligations sit outside the primary corporate balance sheet, routed through operating leases, take-or-pay GPU purchase contracts, and special-purpose-vehicle (SPV) joint ventures ^[1]. The mechanics matter because they change who actually eats a default: when debt is issued through an SPV, it sits on the SPV's books, not the hyperscaler's, which lets a company keep building AI capacity without its own credit rating absorbing the hit. Meta's roughly $30 billion financing for its Hyperion data center in Louisiana is a case in point - structured so the obligation doesn't show up as a direct liability on Meta's own balance sheet ^[3]. Oracle took the opposite path. Rather than routing its data-center buildout through off-balance-sheet vehicles, Oracle loaded the debt directly onto its own books, pushing its on-balance-sheet debt-to-equity ratio toward 500% - and over 10x once hidden obligations are counted - which is exactly why S&P cut its long-term rating to BBB-, just one notch above junk ^[2]. The contrast is instructive: hiding leverage in an SPV buys a company breathing room with rating agencies, but it also makes it much harder for investors to size up the sector's true risk, since the same $1.65 trillion simply isn't visible in the numbers Wall Street is used to scrutinizing.

If the hidden-debt story is abstract, CoreWeave made it concrete. Its five-year credit default swap spread surged above 1,140 basis points before settling near 1,038 bps - a 164% jump year-over-year - with swap pricing implying more than a 60% market-priced probability the company could default within five years, an intraday credit-risk record ^[4]. Its 9.75% bond due 2031 now yields 15.78%, a level that reflects genuine distress pricing rather than routine volatility. The stock fell nearly 12% on a single Friday and about 24% for the week, against a debt and obligations pile exceeding $51 billion ^[5]. What makes CoreWeave the sector's test case rather than just one company's problem is the business model underneath it: unlike the diversified hyperscalers, a neocloud like CoreWeave is financing capex almost entirely through debt and equity issuance rather than operating cash flow, which is precisely the structure Wall Street is now pricing for risk. Online retail-investor discussion has split sharply on what this means - skeptic communities read the CDS spike as the first domino, while more technical threads push back that the same hyperscalers CoreWeave depends on for contracts have real operating revenue of their own and are unlikely to need a bailout even if CoreWeave itself does stumble.

Strip away the company-specific drama and the sector-wide math is the real source of anxiety. Gartner projects AI infrastructure spending will reach roughly $7 trillion through 2030, against AI-related revenue that tops out at no more than $3.6 trillion over the same stretch - a gap the firm says could translate into more than $4 trillion in asset write-downs by 2032, with as many as 80% of neoclouds potentially shutting down before the decade is out ^[6]. That forecast lands against a backdrop where AI capex has already climbed to an estimated 93% of hyperscalers' operating cash flow in 2026, up from just 33% in 2023 ^[8]. In other words, the companies funding this buildout are plowing nearly everything they generate back into infrastructure, with the shortfall increasingly covered by debt rather than cash on hand. Fortune's reporting frames this less as a single trigger event than a slow-building credit story - the kind of warning that has been sounded, and shrugged off, multiple times over the past three years without a crisis arriving ^[7]. The open question isn't whether the spending-revenue gap is real; the data says it is. It's whether AI monetization closes the gap before credit markets lose patience with the companies betting it will.

The most contrarian framing circulating around this selloff argues that AI itself is almost beside the point. On this view, the real risk sits in the bond market: rising Treasury yields, driven by broader government debt-to-GDP dynamics rather than anything specific to AI, raise the cost of capital for every business model that depends on a financing spread - AI infrastructure simply happens to be the most leveraged, most visible example right now. Jamie Dimon's own warning fits this reading closely. Rather than calling AI a bubble, he frames the risk as a competition for capital: AI-driven corporate borrowing is competing directly with U.S. government borrowing for the same pool of money, and he expects that pressure to eventually feed into corporate credit spreads the way it normally does in credit cycles ^[9]. That reframing matters because it shifts the question investors should be asking. If this is fundamentally a capital-markets story, then AI revenue actually arriving on schedule doesn't fully solve the problem - rising rates would still squeeze every AI company's cost of debt regardless of how good its product is. The AI-specific debt data (the $23 billion September issuance, the CoreWeave CDS spike, the Oracle downgrade) may be the most visible symptom of a wider, less AI-specific credit tightening, rather than its sole cause.

What makes this moment unusual is that the disagreement is still completely unresolved in real time. On October 16, the Nasdaq posted its worst week since June on AI debt fears. Three days later, Goldman Sachs was telling clients to buy the CoreWeave dip, arguing 'not all neoclouds are created equal' ^[10]- directly at odds with Gartner's $4 trillion write-down warning issued in the same news cycle. Banks, meanwhile, are reportedly already pulling back from AI-related lending to avoid concentration risk, even as equity desks recommend adding exposure to the same sector ^[11]. That split extends into public sentiment: some retail-investor communities are treating the credit stress as proof the AI boom is unwinding, with a recurring expectation that governments would ultimately treat AI infrastructure as too systemically important to let fail; others point out that this exact 'bubble popping' warning has resurfaced every few months for years without the crash materializing. Both camps are working from the same underlying numbers - the debt issuance collapse, the CoreWeave CDS spike, the spending-revenue gap - and reaching opposite conclusions about what happens next. That is itself the story: the facts on AI debt are no longer in serious dispute, but what they mean for the companies and investors exposed to it still is.