AI's $500 Billion Debt Binge Signals a Structural Repricing of Tech Credit Risk in H2 2026

The bond market is funding the AI buildout on terms that assume nothing breaks. Morgan Stanley now expects roughly $570 billion of AI-related global debt issuance in 2026, with $236 billion already printed by end-May [1]. The second-order effect is not a credit event. It is that passive investment-grade index holders are being turned into involuntary AI infrastructure financiers, and the repricing, when it comes, will run through pension allocations long before it touches a hyperscaler.

The capital structure of AI has changed, and almost no equity analyst is modelling it

Until 2024, the dominant mental model of AI infrastructure was venture-equity logic: write down the capex, trust the operating leverage, judge the multiple. That model is now obsolete for the four companies that matter. Alphabet, Amazon, Microsoft and Meta will spend roughly $700 billion of capex in 2026, with Morgan Stanley projecting the hyperscaler total crosses $1 trillion in 2027 [2]. The five largest tech borrowers sold around $121 billion of US corporate bonds in 2025, more than four times their 2020-2024 average of $28 billion [3].

The size of individual deals tells you what is happening to the term structure. Meta's October 2025 sale was $30 billion, the largest non-merger investment-grade deal on record. Alphabet raised about $31.5 billion in February 2026 and included a 100-year bond. Amazon priced $37 billion across 11 tranches on 10 March 2026, with orders running roughly four times the size on offer [4]. Century bonds and 11-tranche jumbos are not the financing tools of a software company. They are the tools of regulated utilities, sovereigns and pipeline operators. Hyperscalers are accepting that classification, because it is the only way to fund the numbers.

The consequence sits inside the investment-grade index. According to MUFG analysis cited in the Forbes piece, hyperscalers accounted for four of the five largest US high-grade deals of 2025 [5]. Anyone running a passive IG bond mandate, including the median US corporate pension and a long list of insurance general accounts, is now long AI capex whether their investment committee voted on it or not. That has never been true before. It is the first thing corporate treasurers and pension trustees should be asking their fixed-income consultants this quarter.

Spreads price perfection; the equity market has stopped doing so

The credit market is currently underwriting hyperscaler debt on an assumption of permanently high revenue visibility and permanently top-tier balance sheets. Fidelity's Stacie Ware has said plainly that "investors aren't well compensated to own corporate securities right now," and the firm's Total Bond Fund is actively steering away from hyperscaler paper [6]. DoubleLine and Oaktree are now positioned for AI credit pain, buying instruments designed to perform if conditions deteriorate [7]. The cost of insuring hyperscaler debt through credit default swaps has been rising since mid-2025 [8].

The asymmetry is what should worry a corporate treasurer holding this paper. A bond at tight spreads cannot appreciate meaningfully above par. It can lose multiple points on a one-notch downgrade. With the 30-year Treasury above 5% and the 10-year above 4.5%, the duration risk on a 100-year Alphabet bond is brutal in any scenario where rates drift higher or AI revenue growth disappoints by a quarter or two [9].

The equity market is already pricing some of this risk. Broadcom's tepid AI guidance wiped roughly $444 billion of market capitalisation in two days in early June 2026 and dragged the Nasdaq to its worst session in 14 months [10]. Equity has decided that "good but not perfect" AI growth is a sell signal. Credit has not yet repriced that recognition. That gap is the trade DoubleLine and Oaktree are setting up against, and it is the gap a corporate treasurer should be hedging now, not in Q1 2027.

Oracle is the canary; enterprise AI budgets are the cage

Within the hyperscaler cohort, credit dispersion is already starting. Oracle's free cash flow has turned negative because of AI infrastructure commitments and is projected to remain negative through 2030 [11]. Unlike the other four, Oracle does not have a trillion-dollar advertising or consumer cloud business to absorb the capex. Its AI bet is the company. If a single rating agency moves Oracle one notch, the broader market gets a live test of how the IG index handles a concentrated AI issuer under pressure. Any Oracle rating action between now and year-end should be read as a market-wide signal, not a single-name story.

The demand side is also softening in ways that bondholders should be tracking but mostly are not. Uber reportedly exhausted its 2026 AI coding budget by April. Microsoft revoked Claude Code licences within months of issuing them. The FinOps Foundation has noted that many enterprises spent three times their planned annual AI budget by April 2026 [12]. The bull case for hyperscaler credit assumes enterprise AI spend keeps compounding. The early evidence is that customers are starting to ration. If 2027 enterprise AI budgets are flat rather than sharply higher, the revenue projections backing the IG ratings on this paper get a serious second look.

Morgan Stanley itself notes that chip company AI financing is migrating toward shorter, fully-amortising structures rather than long-duration bonds [13]. Credit markets are already differentiating within the AI supply chain. The implication is that any cliff edge will not arrive evenly. It will concentrate in the issuers who took the longest duration at the tightest spreads, which is the hyperscaler cohort.

The counter-case: maturation, not fragility

The serious version of the bull argument is that AI infrastructure has earned its utility-style financing. Alphabet, Amazon, Microsoft and Meta carry top-tier ratings, generate hundreds of billions in operating cash flow, and are funding assets with genuine 20-year economic lives. Barclays expects total US investment-grade issuance to reach $2.46 trillion in 2026 [14]. In that context, even a $570 billion AI slug is roughly a quarter of the market, large but absorbable. Amazon's $37 billion deal drawing four-times oversubscription suggests demand is not just adequate but enthusiastic [15]. DoubleLine's own Robert Cohen has said prices and valuations on bonds "aren't yet frothy" [16]. Morgan Stanley adds that hyperscalers are broadening their investor base through non-USD issuance, distributing supply globally rather than stacking it on US accounts [17].

That argument is real, and on a one-year view it will probably hold. The weakness is that it conflates issuer quality with bondholder outcome. The issuers can remain investment-grade and the bondholders can still lose money. That is the lesson of every previous infrastructure debt cycle from railroads to telecom: the assets get built, the operating companies survive in some form, and the original buyers of the long-duration paper take the hit through downgrades, spread widening and mark-to-market losses. The bull case is right that hyperscalers will pay their coupons. It does not address what the bonds trade at in 2028 if AI revenue compounding slows materially, which is the scenario the equity market is already starting to flirt with after Broadcom.

What to watch

1. Spread differentiation between Oracle and the big four hyperscalers over the next two quarters. If Oracle's IG spread widens materially relative to Alphabet between now and end-Q1 2027 without a downgrade, the market is pricing idiosyncratic AI capex risk inside the IG cohort. If spreads stay tightly clustered, the credit market is still treating hyperscalers as a single block, and the concentration risk in passive IG mandates remains unhedged.

2. The 2027 capex guidance from Alphabet, Amazon, Microsoft and Meta on Q4 2026 earnings calls. Morgan Stanley's $1 trillion 2027 capex projection [18] is the implicit basis for the 2027 bond supply forecast. If any two of the four guide capex flat or down year-on-year, the supply story breaks and outstanding long-duration paper rallies sharply. If all four guide above consensus, the supply pressure intensifies and 30-year and 100-year tranches reprice wider.

3. Enterprise AI budget realisation versus plan for fiscal 2027, as reported by the FinOps Foundation or equivalent surveys by end-March 2027. If enterprise AI spend in 2027 grows only modestly year-on-year, the revenue model underpinning hyperscaler IG ratings comes under serious agency scrutiny. If growth runs sharply above plan, the credit thesis survives another year and the Fidelity caution looks early.