When the Credit Market Builds a Fire Exit
What Four Products in Five Months Tell You.
On Monday, Bloomberg reported that JPMorgan launched a basket of credit default swaps on five hyperscalers — Alphabet, Amazon, Meta, Microsoft, and Oracle — tradeable in $25 million increments.[1] Three days earlier, Meta, Alphabet, and Microsoft were added to the CDX Investment-Grade Index, the 125-company benchmark that defines how institutional investors hedge corporate credit risk.[2] Four months before that, Citadel Securities began making markets in two baskets of hyperscaler bonds — one with ten-year maturities, one with thirty-year maturities.[3]
Twelve months ago, single-name CDS on these companies barely existed. Today, they are among the most actively traded US derivatives contracts outside the financial sector.[4]
The speed matters more than the products. When Wall Street builds hedging infrastructure this fast, it is not reacting to a risk. It is pricing one in.
What the Credit Market Sees
The equity market treats hyperscaler AI spending as a growth story. The credit market treats it as a leverage story. Both are looking at the same balance sheets and reaching different conclusions — and neither is obviously wrong. The hyperscalers collectively hold roughly $350 billion in cash, are projected to generate approximately $725 billion in operating cash flow in 2026, and carry a liabilities-to-assets ratio of 48%, half the S&P 500 average.[5] Saba Capital is selling CDS protection on these names, collecting premiums on what it views as overpriced AI fear.[6] The balance sheets are not telecom balance sheets.
But the volume of debt issuance is approaching telecom-scale levels. The Big Five hyperscalers issued $121 billion in bonds in 2025 — up from an annual average of $28 billion between 2020 and 2024 — with over $90 billion raised in the last three months of the year alone. Alphabet issued a 100-year bond in February 2026.[7] BofA expects the Big Five to borrow roughly $140 billion annually over the next three years, potentially exceeding $300 billion annually — putting them on pace with the Big Six banks’ expected issuance.[8] Barclays projects total US corporate bond issuance will hit $2.46 trillion in 2026, with hyperscaler AI capex as the single largest upside risk.[9]
Aggregate capex for the Big Five, after buybacks and dividends, now exceeds projected operating cash flows.[10] These companies — historically self-funding — are borrowing to sustain the AI buildout. “Chip and Mortar” traced this shift at the company level: Amazon’s infrastructure reversal showed how capex funded by operating cash flow becomes capex funded by debt when the bet gets big enough. That shift from self-funded to debt-funded is what moved the CDS market from dormant to hyperactive in under a year.
Oracle is the canary. Its five-year CDS spread tripled from 43 basis points to over 135 basis points between September and December 2025 — the highest level since the 2008-09 financial crisis.[11] Its total borrowings exceeded $108 billion by late 2025 and have continued to grow, making it the largest non-bank issuer in the Bloomberg US Corporate Index.[12] S&P revised its outlook to negative.[13] In January, bondholders sued, alleging Oracle failed to disclose plans for $38 billion in additional borrowing when it sold $18 billion in bonds the previous September.[14] Oracle’s CDS now averages over $830 million in weekly trading volume — the most liquid investment-grade CDS in the market.[15]
None of this is surprising if you read the balance sheet analysis in “Cloud vs. Clout” or the codependence architecture in “Hotel Abilene” — the five-stage access sequence described there maps precisely onto this covenant pressure stage.
But Oracle is not the structural story. Oracle is the outlier that made the structural story visible. The structural story is that the entire hyperscaler credit complex is being repriced as a distinct risk category.
The Product Sequence Is the Signal
Three products in five months, each escalating the tradability of hyperscaler credit risk, with a fourth building alongside:
November 2025: Citadel Securities’ bond baskets. Cash bond positions in four hyperscalers at two durations.
February 2026: JPMorgan’s CDS basket (reported publicly today). Synthetic exposure to five hyperscalers in $25 million clips. No bond ownership required.
March 2026: CDX IG Index inclusion — a mechanical semi-annual reshuffle, but one that now embeds Meta, Alphabet, and Microsoft in the benchmark every institutional portfolio uses to manage credit risk.
Some of this is market infrastructure catching up with market size. But when BofA’s chief investment strategist calls hyperscaler bonds the best short, and banks are approaching hedge funds for protection for the first time, the infrastructure is being used directionally. JPMorgan isn't alone. Goldman Sachs is reportedly pitching a parallel product that gives hedge funds the same capability — a way to bet on whether AI debt holds up.[16]
The BIS noticed. In a March 2026 report, officials flagged what they called “shadow borrowing”: hyperscalers using off-balance-sheet arrangements to finance data center expansions with private credit firms—obligations that are economically equivalent to debt but are outside corporate balance sheets, creating new shock transmission channels.[17] If you read “Compute Equals Commitments,” the gap between announced AI investment and actual capital expenditure was already the market’s blind spot. The BIS just named the next one: the gap between on-balance-sheet debt and total economic obligation.
This Has Happened Before
The last time Wall Street built hedging infrastructure this fast for a single sector was the telecom boom. Between 1996 and 2001, real private fixed investment in communications equipment more than doubled — from $62 billion to over $135 billion annually — much of it debt-financed.[18] Telecom became the largest sector in the Merrill Lynch High-Yield Bond Index, accounting for 20.3% of the index by 2000.[19] Credit markets saw the problem first: telecom junk bond spreads rose 100–130% between late 1997 and late 1998, while telecom equity investors were still earning 18% annualized returns.[20] The CDS market grew from $151 billion in notional value in 1997 to $2 trillion by the end of 2002, driven primarily by banks hedging telecom loan exposure.[21] By mid-2002, $110 billion in telecom bankruptcies had been filed — a quarter of all corporate defaults.[22]
The pattern repeated before 2008. The ABX.HE index — essentially insurance contracts on baskets of subprime mortgage bonds, packaged into a tradeable index — launched in January 2006, created by dealers to make the subprime market more liquid.[23] Michael Burry had closed his first CDS trade on subprime eighteen months earlier — he had to persuade banks to create the instruments.[24] The dealers built the infrastructure; the informed shorts were the first to use it. The hedging infrastructure was in place 18–36 months before the crisis. Goldman Sachs was marketing ABX shorts to hedge funds by summer 2006.[25] Bear Stearns CDS spreads remained near the risk-free rate until four months before the firm collapsed.[26]
The consistent pattern across both cycles: massive debt accumulation forces credit markets to build new risk-transfer infrastructure. Sophisticated participants begin using it. Defaults arrive 18–36 months later, while less-informed participants are still positioned long. Not every hedging infrastructure buildout precedes a crash — energy CDS grew rapidly in 2014-15 and the sector restructured without systemic failure. The signal is not deterministic. But the signal was never what existing instruments priced — it was that entirely new product categories were created to hedge previously unhedgeable risks.
What This Means
If you manage a portfolio, you now have hyperscaler AI credit exposure whether you chose it or not. The CDX IG inclusion embeds it in every index-level credit hedge. The BIS shadow borrowing finding suggests private credit and insurance allocations may carry it as well.
The question is not whether hyperscaler AI capex produces returns. It does. The question is whether your portfolio’s exposure to the downside scenario — where AI revenue disappoints and $600 billion in annual capex meets $140 billion in annual debt issuance meets extended depreciation schedules that flatter the income statement — is a position you chose or one you inherited.
JPMorgan built a basket so its clients could answer that question. The fact that the basket exists is the answer to a different question: whether the credit market thinks the AI capex cycle carries tail risk worth hedging.
It does. The fire exit is now open. That tells you something about what the building inspectors found.
Notes
[1] Bloomberg, “JPMorgan Offers Clients a New Way to Hedge AI Debt Risk,” March 23, 2026. Basket includes CDS on Alphabet, Amazon, Meta, Microsoft, and Oracle; trades in $25 million increments ($5 million per firm).
[2] Bloomberg, “Meta, Alphabet Join Credit-Risk Index as AI Hedging Demand Soars,” March 20, 2026. See also GuruFocus/Yahoo Finance. Meta, Alphabet, and Microsoft added to S&P Dow Jones Indices’ CDX Investment-Grade Index effective March 20. The index is equally weighted across 125 names; each hyperscaler represents ~0.8% of the index. The exposure is dilute per-name but cumulative across all hyperscaler constituents.
[3] Bloomberg, “Citadel Securities Launches AI Bond Trading Baskets for Hedging,” November 19, 2025. Hedgeweek confirmed baskets include 10-year and 30-year maturities of bonds issued by Microsoft, Amazon, Alphabet, and Meta.
[4] DTCC data, as reported by Bloomberg, March 23, 2026, and S&P Dow Jones Indices.
[5] JPMorgan, as cited by FT, November 2025: hyperscalers collectively hold ~$350 billion in liquid cash and investments and are projected to generate ~$725 billion in operating cash flow in 2026. CreditSights, Q3 2025: hyperscaler liabilities-to-assets ratio fell to 48%, close to 2015 levels, versus S&P 500 steady at ~80%. Per MUFG Americas, December 2025.
[6] Reuters exclusive, November 17, 2025, by Nell Mackenzie and Lucy Raitano. Saba Capital (Boaz Weinstein) selling CDS protection on Oracle, Microsoft, Meta, Amazon, and Alphabet to banks seeking to hedge AI-related credit exposure. Source with direct knowledge of the trades. First time banks had approached Saba for this specific trade. BofA CIS Michael Hartnett, same week: “Best short is AI hyperscaler corporate bonds.” Confirmed by Hedgeweek, November 19, 2025.
[7] Mellon Investments, “Record-Breaking AI-Related Debt Issuance in 2025,” citing Bloomberg data as of December 15, 2025: hyperscalers issued $121 billion in bonds in 2025, with over $90 billion in the last three months alone, up from a $28 billion annual average 2020–2024. BIS Quarterly Review, March 2026 confirms issuance “topped $100 billion.” Alphabet 100-year sterling bond (£1 billion) priced February 10, 2026: Bloomberg, Reuters, CNN. First century bond by a tech company since Motorola in 1997. Part of $20 billion multi-tranche dollar offering upsized from $15 billion after >$100 billion in orders.
[8] Reuters, “AI hyperscalers will drive higher US corporate bond supply in 2026, analysts say,” January 15, 2026. BofA estimate: Big Five to borrow ~$140 billion annually over three years, potentially exceeding $300 billion annually.
[9] Barclays report, January 2026, as reported by Reuters. Barclays projects $2.46 trillion in total US corporate bond issuance (up 11.8% from $2.2 trillion in 2025) and $945 billion in net issuance (up 30.2% from $726 billion).
[10] MUFG Americas, “AI Chart Weekly: Financing the AI Supercycle,” December 19, 2025. Aggregate capex for the Big Five, after buybacks and dividends, exceeds projected cash flows. CreditSights 2026E data.
[11] BondBloX, January 2026, citing Bloomberg data. Oracle 5Y CDS surged from 43 bps to its highest level since the 2008–09 financial crisis. ROIC.ai reported 139 bps on December 11, 2025, citing S&P Global Market Intelligence.
[12] BondBloX, January 2026: approximately $105 billion in total debt, including $95 billion in corporate bonds. ainvest.com (December 2025) cites $108 billion. Winbuzzer (citing Fortune, March 24, 2026) reports $124 billion in current borrowings — single source, not independently confirmed. Growth trajectory is directionally confirmed across all three measurements.
[13] S&P revised Oracle’s outlook to negative in July 2025 on expectations of weakening cash flows, per BondBloX. Moody’s described Oracle as having the weakest credit metrics among investment-grade hyperscalers.
[14] CNBC, Reuters, Bloomberg, January 14–15, 2026. Ohio Carpenters’ Pension Plan v. Oracle Corp., filed in New York state court. Bondholders allege Oracle failed to disclose plans for $38 billion in additional borrowing when selling $18 billion in bonds on September 25, 2025.
[15] S&P Dow Jones Indices, citing DTCC data, as reported by Bloomberg. Oracle CDS averages over $830 million in weekly trading volume — the most liquid investment-grade CDS.
[16] Goldman Sachs reportedly pitching hedge funds total return swaps on corporate loan price swings. Winbuzzer, “JPMorgan Launches CDS Basket to Hedge AI Debt Risk,” March 24, 2026, citing Fortune. Single secondary source; not independently confirmed at time of publication.
[17] BIS Quarterly Review, March 2026. Authors: Egemen Eren, Ingomar Krohn, Karamfil Todorov. Structures involve dedicated vehicles and special-purpose entities capitalised with equity from consortia and raising debt through private placements. Banks support vehicles with funding lines, creating links to insurers and private credit vehicles, and new refinancing pressure and guarantee activation channels.
[18] Richmond Fed Economic Quarterly, Fall 2003, Alexander L. Wolman, “Boom and Bust in Telecommunications.” Real private fixed investment in communications equipment surged from $62 billion annually (1996) to over $135 billion (Q4 2000).
[19] Edward Altman and Gaurav Bana, “Defaults and Returns on High Yield Bonds: The Year 2002 in Review and the Market Outlook,” NYU Stern Working Paper, 2003. Telecom accounted for 20.3% of the Merrill Lynch High-Yield Bond Index and 52% of all defaulted dollar amounts in 2002.
[20] San Francisco Fed Economic Letter 2001-33, “Rising Junk Bond Yields: Liquidity or Credit Concerns?” November 2001. Telecom junk bond spreads rose 100–130% from November 1997 to October 1998 while annualized stock returns for telecom junk issuers averaged +17.8%.
[21] ISDA Market Survey, year-end 2002. CDS notional grew from $151 billion (mid-year 1997 survey) to $2 trillion (year-end 2002 survey). ISDA credited credit derivatives with having prevented severe distress from telecom bankruptcies. CDS documentation standardized by ISDA in 1999.
[22] FCC Chairman Michael Powell, testimony to Senate Commerce Committee, July 2002. $110 billion in telecom bankruptcies filed over the preceding 18 months. Overall high-yield default rate hit 12.8% in 2002 (Altman, fn 19).
[23] ABX.HE index launched January 19, 2006, administered by Markit Group. Created by a consortium of 16 dealer desks. Reserve Bank of Australia, “Box B: The ABX.HE Credit Default Swap Indices,” Financial Stability Review, March 2008.
[24] Michael Burry closed first CDS trade on subprime MBS with Deutsche Bank on May 19, 2005, for $60 million, after five banks told him they had no idea what he was talking about. Gregory Zuckerman, The Greatest Trade Ever (2009). Burry accumulated over $1 billion in CDS on subprime by October 2005.
[25] Goldman Sachs Mortgage Department developed “predominantly pessimistic view” by mid-2006. Senate Permanent Subcommittee on Investigations, “Wall Street and the Financial Crisis: Anatomy of a Financial Collapse,” April 2011. Also available via FRASER/St. Louis Fed.
[26] Bear Stearns CDS spreads were statistically indistinguishable from the risk-free rate before November 2007; the firm collapsed in March 2008. Trabelsi and Hammami, “Tail return analysis of Bear Stearns’ credit default swaps,” Economic Modelling, 2011.