NextFin News - The insurance market on Big Tech debt is flashing its loudest warning since the 2008 financial crisis, and it is not coming from equity traders. Credit default swaps referencing the companies bankrolling the artificial-intelligence buildout have surged to record levels, with Oracle's five-year spreads above 200 basis points and net notional protection on major technology names hitting a record $12.5 billion - up roughly 500 percent since the second quarter of 2025. The surge is the credit market's first real pushback against a capex boom that consensus expects to reach $690 billion this year and $870 billion in 2027, up from about $480 billion at the start of 2026. But the warning comes with a catch: much of the demand for protection is mechanical, driven by bank balance-sheet rules rather than a conviction that the hyperscalers are about to default.
The central question is whether the CDS market is pricing genuine credit deterioration or simply absorbing a flood of new debt. The answer matters because it determines whether this is a cyclical air pocket in an otherwise sound credit cycle or the first crack in the financial architecture of the AI boom. The evidence points to both: a cyclical capex super-cycle that will mean-revert if AI monetization disappoints, layered on top of a structural shift in how investment-grade credit risk is intermediated - one that moves concentration risk off bank balance sheets and into less-regulated corners of the financial system.
The Numbers Behind the Surge
The scale of the move is unambiguous. Amazon, Alphabet, Meta and Oracle issued around $195 billion in bonds in the first half of 2026, up about 80 percent from roughly $108 billion across all of 2025, according to an analysis of LSEG data. That issuance wave has a clear destination: Goldman Sachs Research estimates nearly $500 billion of AI-related debt issuance so far in 2026, and projects cumulative hyperscaler capital expenditure of $1.15 trillion from 2025 through 2027.
Oracle is the outlier that makes the aggregate story legible. Its five-year CDS spread has widened to more than 200 basis points, a record since the 2008 financial crisis, from about 40 basis points a year ago, after S&P Global Ratings cut the company to BBB-, one notch above junk. The company carries roughly $130 billion of debt and posted negative free cash flow of $23.7 billion for fiscal 2026, with a debt-to-equity ratio near 4x - compared with 0.30x for Microsoft, 0.36x for Meta, 0.18x for Alphabet and 0.51x for Amazon. Oracle's 2026 capital expenditure alone was $55.7 billion. Unsurprisingly, Oracle accounts for about $6.5 billion of the record $12.5 billion net notional in CDS on major tech names, with Amazon and Alphabet at roughly $2 billion each.
Excluding Oracle, the picture is calmer but still elevated. Market data show five-year CDS spreads for Amazon, Alphabet and Microsoft have risen to approximately 49 basis points, their highest levels since 2018. Apollo Global Management's chief economist Torsten Slok notes that five-year CDS spreads on hyperscalers have surged more than 50 basis points since the start of 2025 to around 75 basis points - about 20 basis points above levels seen during the 2022 bear market. AI-related debt, which was one percent of investment-grade supply in 2024, has been 18 percent of supply in the first half of 2026. And the gross leverage of the hyperscalers has doubled from 0.9x to 1.8x in just six months, now sitting above the entire energy sector's leverage.
The market is also inventing new plumbing to handle the flow. JPMorgan introduced a standardized basket product covering the five major hyperscaler names in $25 million increments, offering institutional clients a more liquid hedging vehicle than single-name instruments. Meta and Alphabet only entered CDS markets as reference entities in November 2025. Monthly notional volumes of hyperscaler CDS trading at Bank of America increased tenfold in the first months of 2025 relative to the period before it. CDS contracts referencing Microsoft, Amazon and Oracle recorded $4.6 billion in notional trading volume in the first quarter of 2026, against $759 million in the equivalent period of the prior year.
Why This Is Different From a Normal Credit Cycle
A normal credit cycle widens because borrowers deteriorate. This one is widening partly because the intermediaries are full. The most important detail in the entire story is not the level of spreads but who is buying the protection and why. Bank of America's head of US single-name CDS has confirmed publicly that much of the demand for hyperscaler CDS originates with credit valuation adjustment, or CVA, desks, which are purchasing protection with the express aim of preserving lending capacity within regulatory concentration limits.
The mechanism runs like this. Banks face regulatory ceilings on how much exposure they may carry against a single counterparty across loans, derivatives and other instruments. As hyperscaler borrowing accelerated - the five major AI hyperscalers went from issuing an average of $28 billion a year in US corporate bonds between 2020 and 2024 to $121 billion in 2025 alone - major banks began approaching those limits. The CDS becomes the release valve: buy protection, reduce the risk-weighted exposure, keep lending. The demand is balance-sheet-driven, with the management of regulatory constraints rather than fundamental credit assessment as the operative driver.
That distinction matters because it means spread width is no longer a clean read on default probability. It also means the risk has not disappeared; it has moved. When CVA desks systematically purchase protection from hedge funds and other non-bank financial intermediaries, concentration risk is transferred to institutions not subject to equivalent prudential oversight, with no corresponding mechanism for tracking the aggregate magnitude of that transfer. The structural incentive to hedge is preserved; the systemic visibility required to assess the consequences of that hedging is absent. The Basel III CVA frameworks were designed to ensure banks hold adequate capital against derivatives exposures; they were not designed for a scenario in which the same small group of investment-grade counterparties simultaneously dominates both corporate lending books and derivatives counterparty relationships at a system-wide level.
This is the structural leg of the call: the CDS market has been permanently rewired around AI credit. Investment-grade technology names, long considered too safe to need credit insurance, are now reference entities with liquid single-name and basket markets. That architecture will not unwind when the capex cycle turns.
The Cyclical Leg: A Capex Boom That Must Be Paid For
The cyclical leg is the capex cycle itself, and it bears an uncomfortable resemblance to previous investment booms. Investment in equipment and software as a share of US GDP is now on track to surpass the peak reached in the late 1990s. Hyperscaler capex growth relative to starting total assets is projected to outpace that of telecom in the late 1990s, although it is still projected to be lower as a share of operating cash flows. Torsten Slok of Apollo puts the pace in perspective: consensus expects hyperscaler capex at 3 percent of GDP every single year for the next three years, and a cycle that builds at 0.85 percentage points a year can unwind at a similar pace. That, rather than the buildout itself, is the macro risk if AI demand disappoints.
"So far this year, again, we've had $194 billion of issuance. We think that roughly 33% or one-third of CapEx will be debt financed in 2026. So, what that means is that all in, we're bracing for something like $250 billion of direct supply from the hyperscalers, leaving the project finance aside," said Zach Ablon, head of the credit sales desk in Global Banking & Markets at Goldman Sachs.
The debt-financed share is the vulnerable seam. Cash-rich technology giants are turning to bond markets because the scale of the buildout has outrun even their formidable free cash flow. Future lease obligations add another layer: recent 10-Q filings point to a combined $822 billion of future non-discounted lease commitments, versus $675 billion as of the end of February 2026, that have yet to be recognized on hyperscaler balance sheets. Of the $969 billion in total commitments across the group, $662 billion is in data centre leases yet to commence. When those leases hit the income statement, they will test whether AI revenue can cover both the capex and the rent.
The Counter-Thesis: It Is Hedging, Not a Default Call
The strongest argument against reading the CDS surge as a credit alarm is that the market for these contracts remains tiny relative to the debt being hedged. According to Depository Trust & Clearing Corporation figures, the combined average daily notional volume of CDS contracts on 16 technology firms traded or cleared in the second quarter was $637.5 million - barely 4 percent of the $16 billion average daily notional volume across all corporate and sovereign CDS contracts. The average number of trades per day in technology CDS was 54, with exactly one-third of that in Oracle. Fourteen of the remaining names saw single-digit average daily trades, most fewer than five, while Apple's CDS did not trade at all.
In other words, a handful of names dominate a thin market, and thin markets move on flow. These spreads are not necessarily widening because investors believe these companies will default, but because CDS is a relatively straightforward and cheap vehicle to hedge against increasing exposure to the AI sector. And the implied probabilities, while rising, are not at distress levels. Strategists at Societe Generale calculate the cumulative probability of default implied by five-year CDS spreads for U.S. hyperscalers at around 7 percent - a sharp increase since last year and now higher than the broader U.S. investment-grade average of around 4.5 percent. Oracle pulls that aggregate figure materially higher, with an implied default probability above 16 percent. These companies have never defaulted on their debt obligations. Distressed credits trade in the hundreds or thousands of basis points; Oracle's 200-plus is elevated for investment grade but not in that territory.
This counter-thesis has force, but it rests on a premise that the next two quarters will test: that issuance can keep flowing smoothly and that free cash flow will recover. If either breaks, the mechanical-hedging explanation stops being sufficient.
What Would Prove the Read Wrong
The falsifying signal is specific. If Oracle, Amazon and Alphabet five-year CDS spreads sustain levels above 250 basis points while hyperscaler free cash flow stays negative for four or more consecutive quarters and annual capex remains above $650 billion, the mechanical-hedging thesis fails and fundamental credit repricing is underway. A second trigger would be a failed hyperscaler bond auction, or a downgrade of any major hyperscaler to BB, junk territory. A softer signal to watch: the share of AI-related debt in investment-grade supply. If it keeps climbing past 18 percent while spreads keep widening, the market is asking for more compensation, not just more hedging.
Who Benefits, Who Is Exposed
The asymmetry is not where the headlines point. The hyperscalers themselves are the least exposed to a CDS-driven repricing in the near term: they are cash-generating machines with access to the deepest capital markets in the world, and several still carry AA ratings. The group most exposed if spreads keep widening is the second tier of AI-adjacent issuers, where the equity story and the coupon story have already started to diverge. Lambda's $917 million leveraged loan to fund a Nvidia chip lease is one live data point from that tier. So is the project-finance and data-center transaction market, which Goldman Sachs expects to reach $300 billion in 2027 above and beyond direct hyperscaler issuance.
Three scenarios frame the path from here, each with its own trigger. The base case is that spreads stabilize in the 60-to-90 basis-point range for the group as issuance normalizes and free cash flow recovers toward breakeven - the mechanical-hedging thesis holds, and the repricing remains a funding-cost story rather than a solvency story. The upside case for credit arrives if AI revenue per dollar of capex accelerates faster than expected, pulling implied default probabilities back toward the 4.5 percent investment-grade average; that would require hyperscaler free cash flow to turn positive while capex stays above $600 billion a year. The downside case is a credit-event repricing: if Oracle, Amazon and Alphabet five-year CDS spreads sustain levels above 250 basis points while free cash flow stays negative for four or more consecutive quarters, the mechanical explanation fails and the market begins pricing genuine deterioration. The trigger to watch is not the next chip order - it is the next quarterly cash-flow statement.
For investors, the split by time horizon is clean. In the short term, spreads will whipsaw around the issuance calendar and quarterly free-cash-flow prints. In the medium term, the test is monetization: whether AI revenue per dollar of capex can cover the debt service on a $1.15 trillion buildout. In the long term, the architecture has changed permanently - credit derivatives are now a mainstream tool for intermediating investment-grade technology risk, and the concentration risk they create has migrated to institutions outside the prudential perimeter.
The AI boom was financed first by cash, then by equity, and now increasingly by debt. Credit default swaps are the bill collectors arriving early - not because the bill is due tomorrow, but because someone has to check whether it can be paid at all. The market's verdict so far is not default; it is a demand for proof that the buildout earns its keep. That proof will come in free cash flow, not in chip shipments.
Explore more exclusive insights at nextfin.ai.
