NextFin News - Artificial intelligence is no longer just a valuation story; it is becoming a credit story, and the investors best positioned to spot the turn are sounding the alarm. In Bank of America's August Global Fund Manager Survey, 38% of respondents named AI hyperscaler capital spending as the most likely source of a systemic credit event - up from 34% in May and double the share recorded in April. The same survey put an "AI bubble" at the top of the list of tail risks facing markets, cited by 32% of managers. The warning is striking because it comes from investors who are, on balance, deeply bullish on risk assets: cash levels have fallen to 3.5% of assets under management, the sixth-lowest reading since the survey began in 1998, and global equity allocations reached a net 56% overweight, the highest since November 2021.
The shift in sentiment has been rapid. Just four months earlier, AI hyperscaler spending was a fringe concern; today it is the leading candidate for the next credit accident in the view of the largest share of global fund managers. Bank of America strategists led by Michael Hartnett captured the tension in a research note titled "Slow down, hyperscalers," attributing the jump in concern to the magnitude and financing of the AI capital-expenditure boom. The central question the market is now asking is not whether AI will work, but whether the debt used to build it will be repaid.
The Scale of the Buildout
The capital being deployed is of a size that has no clean historical parallel. The five largest hyperscalers - Amazon, Alphabet, Microsoft, Meta and Oracle - are on track to spend roughly $602 billion on capital expenditure in 2026, a 36% jump from 2025, according to CreditSights. That follows a 73% increase in 2025, when the group spent about $443 billion, up from roughly $256 billion in 2024. About three-quarters of the 2026 total, or roughly $450 billion, is earmarked for AI infrastructure: GPUs, servers, networking equipment and data centers rather than traditional cloud capacity.
Company-level commitments make the aggregate concrete. Amazon has guided to $200 billion in 2026 capex; Alphabet to a range of $175 billion to $185 billion; Meta to between $115 billion and $135 billion; Microsoft is tracking toward $120 billion or more after spending $37.5 billion in its most recent quarter alone; and Oracle is targeting roughly $50 billion. Collectively, the five have signaled plans to add about $2 trillion of AI-related assets to their balance sheets by 2030. Capital intensity has reached levels analysts describe as historically unthinkable - Oracle's capital expenditure hit 57% of revenue in the most recent quarter and Microsoft's reached 45%, ratios more typical of utilities or telecommunications networks than of asset-light software companies.
How the Buildout Is Financed
The defining change in 2025 and 2026 is that internal free cash flow no longer covers the bill. For years, hyperscalers funded expansion largely from operating cash flow, which kept credit markets at arm's length from the AI trade. That model has broken down. Goldman Sachs estimates that AI-linked issuers accounted for $141 billion of corporate credit issuance in 2025, already eclipsing the full-year 2024 gross supply of $127 billion. Through the first half of 2026, the same hyperscalers had already issued $194 billion of investment-grade-rated debt, with full-year issuance expected to approach $250 billion - roughly a third of capex.
The borrowing has come in waves. In September and October 2025, AI-focused Big Tech issued $75 billion of U.S. investment-grade debt, more than double the sector's average annual issuance of $32 billion between 2015 and 2024. The total included $30 billion from Meta and $18 billion from Oracle. Oracle has also arranged an $18 billion bond package with maturities stretching to 2065 to fund AI cloud capacity, and a $38 billion project-finance loan is backing its new AI facilities. Bank of America's analysis puts Oracle's total debt at nearly $96 billion against quarterly net income of about $3 billion - a coverage profile that leaves little room for rates to rise.
The financing has also migrated into structures that are harder for investors to see. Special-purpose vehicles, sale-leaseback arrangements and private credit facilities now carry a meaningful share of AI infrastructure debt. A $16 billion Oracle-backed data center campus in Michigan is being financed with roughly $14 billion of debt. CoreWeave raised an $8.5 billion GPU-backed facility in the second quarter of 2026. Amazon secured a $17.5 billion loan dedicated to AI data center construction. Meta closed a $29 billion private credit facility with PIMCO and Blue Owl for AI data centers, with a large portion held off the main corporate balance sheet. These structures keep leverage off the consolidated statements, but they do not eliminate the economic obligation.
Why This Is a Credit Risk, Not Just an Equity De-rating
The mechanism is what separates this concern from ordinary technology-cycle skepticism. Hyperscalers are converting short-duration free cash flow into long-lived, rapidly depreciating assets funded partly with debt. AI hardware is estimated to depreciate at roughly 20% a year, which means a $2 trillion asset base implies about $400 billion of annual depreciation charges before a single dollar of incremental revenue is tested against it. If AI-driven revenue falls short of the embedded expectations, the impairment hits equity first, but the debt holders - bond funds, private credit vehicles and the banks that underwrite them - absorb the residual loss. That is the transmission channel from a technology bet to a credit event.
The exposure is not marginal at the macro level either. An 18% year-on-year increase in information-technology capital investment contributed 1.4 percentage points to U.S. GDP growth in the first quarter of 2026, according to Fitch Ratings. The wealth effect from AI-related equity gains has also been supporting consumer spending at a time when underlying momentum is slowing. Fitch, in its Global Risk Outlook for the third quarter, identified a potential AI-related market correction as one of the two main short-term credit risks for the second half of 2026, alongside geopolitical uncertainty in the Middle East. The agency stopped short of predicting a correction; it flagged the vulnerability that comes from credit markets being more intertwined with AI sentiment than at any previous technology cycle.
The second-order risk is where the credit concern becomes systemic. AI-linked debt does not sit only in the portfolios of specialist funds. It flows into widely held investment-grade bond funds, including those in retirement accounts, and into the loan books of banks that underwrite and warehouse the paper. A repricing in AI credit would therefore transmit across asset classes rather than remaining contained within technology equities. This is the interdependence Fitch highlights: capital markets have become sufficiently embedded in AI sentiment that a re-evaluation of the technology's long-run revenue potential could move credit spreads, not just stock prices.
What Makes This Cycle Different
Every technology boom produces a chorus of "this time it's different." The question is whether the financing structure makes this one structurally more fragile. Three features stand out.
First, the borrowers are investment-grade. Amazon, Alphabet, Microsoft and Meta carry top-tier credit ratings, which has allowed them to tap bond markets at relatively contained spreads. That is the bullish case: these are not speculative-grade issuers, and their balance sheets remain liquid. Meta, for example, carries roughly $37 billion of total debt against more than $60 billion in cash. But investment-grade status is not a shield against cash-flow math. Oracle holds nearly $96 billion in debt, according to Bank of America's analysis, and carries the lowest ratings of the group at BBB, versus the AA and AAA tiers of its peers. Interest expense will consume a growing share of earnings if rates do not fall.
Second, underwriting standards at the project level have become aggressive. The Hut 8 Beacon Point transaction in Texas - a $4.25 billion senior secured notes offering, 100% pre-leased under a 15-year triple-net lease at a 95% loan-to-cost ratio - set a record for a high-performance computing data center bond. JPMorgan, which has been active in financing AI infrastructure, notes that debt concentration in AI lending is now "part of the dialogue" among lenders even if concentration limits are not yet binding. When loan-to-cost ratios approach the full cost of a project and the tenant's business model depends on AI monetization that has not yet been proven at scale, the collateral is only as strong as the revenue assumption underneath it.
Third, the revenue test has not yet arrived. The hyperscalers are building capacity against demand that is largely still implied rather than realized. Microsoft has disclosed an $80 billion backlog of Azure orders it cannot fulfill due to power constraints, which bulls cite as proof of demand. But a backlog is a commitment only if the customer's own AI economics work - and a meaningful share of the announced demand comes from a small number of well-funded AI labs whose combined free cash flow remains deeply negative. If AI applications fail to monetize at the pace the infrastructure assumes, the capex does not shrink quickly. The leases, the debt service and the depreciation keep coming due.
The Counter-Thesis: Why This May Not Break
The strongest argument against the credit-risk thesis is also the simplest: the hyperscalers are not marginal borrowers, and 71% of the fund managers surveyed do not expect any of them to cut capital expenditure this year. Top-rated issuers with large cash buffers can refinance through cycles, and the AI buildout is a winner-take-most race in which falling behind on capacity carries a greater risk than overbuilding. Cloud demand is accelerating, and roughly 75% of hyperscaler capex is tied to AI infrastructure that in many cases already has contracted or committed tenants.
There is also a positioning argument. The same August survey that flagged credit risk showed investors rotating into technology, banks and energy, with U.S. equity allocations at a net 27% overweight, the highest since December 2024. Long global semiconductors remained the most crowded trade at 53%, even after falling sharply from 82% the prior month. The market, in other words, is not behaving like a market that expects a credit accident.
These points are substantial, but they do not fully answer the credit question. Investment-grade balance sheets can absorb a lot, yet the combination of high depreciation, debt-funded expansion and unproven revenue creates a specific vulnerability: a simultaneous hit to earnings and credit spreads. If AI revenue disappoints, equity multiples compress and interest coverage deteriorates at the same time. That is precisely the correlation that turns a sector correction into a credit event. The counter-thesis holds only if revenue catches up to capacity before debt service becomes a constraint - a timing bet, not a structural guarantee.
Who Benefits and Who Is Exposed
The credit buildout creates clear asymmetries across the market. The beneficiaries of continued debt issuance are the lenders and intermediaries: investment banks underwriting the bonds, private credit managers such as PIMCO and Blue Owl, and the bond funds that earn fees on growing assets. Data center developers, utilities and equipment suppliers benefit from the construction cycle regardless of whether the AI economics ultimately work out. These are the parties whose revenue is earned up front.
The exposed parties are the holders of the debt paper and, ultimately, the savers whose retirement accounts own investment-grade bond funds with concentrated AI exposure. If a correction occurs, the losses would not be confined to technology specialists. The interdependence Fitch describes means credit spreads could widen across the investment-grade complex, raising borrowing costs for companies with no connection to AI whatsoever. That is the systemic element - not that hyperscalers will default, but that a repricing of AI credit could tighten financial conditions for the entire economy.
What to Watch
The falsifying signal for the credit-risk thesis is concrete: if the hyperscalers collectively report AI-related revenue growth that matches or exceeds the pace of capex growth for two consecutive quarters, the overinvestment narrative breaks down. Conversely, if capital expenditure keeps rising while AI revenue growth decelerates and interest coverage at the most leveraged name - Oracle, with its BBB rating and debt load approaching $100 billion - tightens below historical norms, the credit concern moves from survey sentiment to realized stress.
Three time horizons matter. In the short term, sentiment and positioning dominate; with cash at 3.5% and equity allocations elevated, a negative catalyst could trigger a sharp rotation. The August survey's contrarian recommendations - long bonds and short commodities, long staples and short tech, long U.K. equities against U.S. - already position for that. Over the medium term, the earnings reports of the hyperscalers will determine whether the capex is being absorbed. Over the long term, the structural question is whether AI becomes a general-purpose technology that justifies a multi-trillion-dollar asset base, or whether the industry repeats the pattern of previous technology cycles in which capacity was built ahead of demand and then written down.
The base case is that the hyperscalers muddle through - refinancing debt, leaning on investment-grade status, and gradually bringing revenue toward capacity. The downside case is a synchronized disappointment in AI monetization that hits earnings and spreads simultaneously. The upside case is that AI revenue accelerates faster than the bears expect, validating the buildout and leaving credit markets unscathed.
"Consensus conviction is no macro landing, no Fed hike, no AI capex cut, no DEM sweep, no bears," Bank of America strategists wrote in the August note. "Positioning continues to recommend investors retreat or rotate within risk assets rather than reload."
The irony is hard to miss: the same investors who are most bullish on the economy are also the ones ringing the bell on the credit risk embedded in that optimism. The AI boom has moved from a story about earnings growth to a story about financing - and financing is where bubbles tend to break.
Explore more exclusive insights at nextfin.ai.
