NextFin News - The rush to borrow hundreds of billions of dollars to finance artificial intelligence is rattling the more than $10 trillion US corporate bond market, forcing a broad repricing of technology credit risk that showed up on Thursday in spiking prices for credit insurance, a selloff in AI-linked equities, and widening spreads across investment-grade and high-yield debt. Oracle Corp., Broadcom Inc. and SpaceX are among the borrowers that priced nearly half a trillion dollars of new debt this year to pay for the computing infrastructure powering AI, and the tally is widely expected to rise by orders of magnitude in the months and years ahead — with Broadcom alone potentially raising about $600 billion to finance computing power in the coming years.
The market's verdict arrived quickly. The tech-heavy Nasdaq Composite fell 1.3% and the S&P 500 dropped 0.5% on Thursday, while the iShares Semiconductor ETF slid 3.5% and Nvidia, the worst-performing Dow component, lost nearly 3%. The repricing is not confined to stocks: five-year credit-default swaps on Oracle reached about 75 basis points, the highest level in at least seven years, and excluding Oracle, the CDS spreads of Amazon.com, Alphabet and Microsoft climbed to roughly 49 basis points, their highest since 2018. Big Tech's credit-insurance costs have more than doubled since early 2025.
The central question this repricing forces investors to answer is whether the AI buildout is running into a cyclical funding hiccup — the kind that mean-reverts once supply digests — or a structural break in how cheaply the world's most valuable companies can finance their ambitions. The evidence points to structural, and that distinction is what makes this selloff different from the volatility investors have learned to buy over the past two years.
The Supply Shock: A Half-Trillion Dollars and Still Counting
The scale of the borrowing is the starting point. Total capital raised by hyperscalers, data-center special-purpose vehicles and neo-clouds — in the form of investment-grade debt, high-yield debt and equity — rose from $172 billion in 2025 to $346 billion so far in 2026, an increase of more than 100%, according to research from Barclays strategist Venu Krishna.
"Over the past two quarters, with unprecedented credit supply hitting the market, credit spreads have widened materially for investment grade and high yield bonds, as well as convertible bonds, across hyperscalers, data centers, and neo-clouds. Along with it, credit default swap spreads have risen sharply for all hyperscalers in 2026."
The concentration among the largest borrowers is striking. The five major AI hyperscalers — Amazon, Alphabet's Google, Meta, Microsoft and Oracle — issued $121 billion in US corporate bonds last year, compared with an average of $28 billion a year between 2020 and 2024, according to a report by BofA Securities. The pace has accelerated: Oracle sold $18 billion of bonds in September, followed in October by Meta's $30 billion deal, the largest-ever individual non-merger high-grade bond sale, with Alphabet and Amazon adding $17.5 billion and $15 billion respectively in November. BofA analysts expect the Big Five to borrow roughly $140 billion annually over the next three years, a pace that could exceed $300 billion a year and put the tech giants on par with the Big Six banks' expected average issuance of $157 billion annually.
Barclays forecasts overall US corporate bond issuance reaching $2.46 trillion in 2026, up 11.8% from $2.2 trillion in 2025, with net issuance of $945 billion, up 30.2% from $726 billion. The point is not that these companies cannot borrow — investment-grade issuers with their ratings can always find buyers. The point is that they are now competing for a finite pool of fixed-income capital at a time when the risk-free rate sits near its highest level in two decades, and every incremental dollar of supply must be absorbed by investors who are already stretched.
The Transmission Mechanism: Why CDS Moves Before Earnings
Credit-default swaps are the canary because they price default risk and balance-sheet leverage directly, without the growth-option premium that inflates technology equities. A spread of 75 basis points on Oracle means it costs about $75,000 a year to insure $10 million of the company's bonds — elevated for a company with Oracle's credit profile, but still far below the hundreds or thousands of basis points that mark genuine distress. The signal is not "default is imminent." The signal is "leverage is rising faster than the market underwrote."
The mechanism runs through three channels. First, the supply channel: a sudden doubling of issuance from one corner of the investment-grade universe widens spreads simply because buyers demand a larger concession to hold incremental duration and credit risk. Second, the leverage channel: debt-financed capital expenditure raises financial leverage even when operating cash flow is strong, and CDS markets reprice that leverage immediately. Third, the discount-rate channel: as the 10-year Treasury yield climbed toward 5.37% earlier in the week — its highest level in 24 years — the denominator in every discounted-cash-flow model rose, compressing the present value of AI payoffs that may be a decade away.
That is why the equity market moved before any earnings disappointment. The selloff accelerated in afternoon trading after a report citing financial documents shared with investors said OpenAI's annualized revenue was about $20 billion lower than previously signaled — a gap large enough to make investors question whether the revenue assumptions embedded in hundreds of billions of infrastructure spending can be met. Roundhill's Memory ETF fell 5%, the semiconductor ETF 3.5%, and the Magnificent Seven tracker 1%.
"The winner of the AI model race keeps looking like whoever owns the distribution toll booth, which is Amazon, Google, Microsoft, Meta and Palantir,"said Shay Boloor, chief market strategist at Futurum Equities, in written commentary — a pointed reminder that infrastructure builders may not capture the economics they are borrowing to construct.
Cyclical or Structural: Why the Supply Wave Does Not Just Roll Over
Investors who have bought every tech dip since 2023 are treating this as cyclical: spreads widen, supply digests, the Fed eventually cuts, and the AI trade resumes. That view has a real case. Credit spreads are mean-reverting by nature; the companies involved are not distressed; and if AI revenue materializes, today's leverage will look conservative in hindsight. A modest widening in CDS is exactly what a functioning credit market should produce when borrowers add leverage — not a distress signal, but a price.
But this episode differs in three structural ways that make a clean mean-reversion less likely, and they are worth separating from the cyclical noise rather than blending together.
First, the supply is not a one-off refinancing wave — it is a multi-year capital program. Broadcom's potential $600 billion raise is not debt being rolled; it is new money funding computing power that does not yet produce commensurate cash flow. When issuance is tied to a multi-year buildout rather than a refinancing cycle, the supply pressure persists through multiple Fed meetings, not just one. That is the difference between a wave that rolls over and a tide that stays higher.
Second, the borrowing is happening at the top of an interest-rate cycle, not the bottom. Companies that locked in sub-2% debt in 2020 and 2021 face no refinancing wall for years. The AI borrowers are issuing at 5%-plus yields into a market where the risk-free rate has just touched a 24-year high. That timing mismatch is structural: it cannot be undone by a single rate cut, and it raises the hurdle rate for every AI project that depends on cheap leverage to clear.
Third, the governance friction has begun. Bondholders sued Oracle this week, alleging the company chaired by Larry Ellison failed to disclose that it would need to sell significant additional debt to build out its AI infrastructure. Legal challenges of this kind appear when capital plans outrun disclosure — and they do not mean-revert with the next earnings print. Taken together, these three factors argue that the repricing is structural at the margin: the floor under the cost of AI capital has moved up, and it will not move back down on its own.
The Counter-Thesis: The Market Is Overreading a Funding Success
The strongest argument against the structural-repricing thesis is also the simplest: this debt is being issued because the borrowers can afford it. Oracle, Microsoft, Alphabet and Meta generate enormous free cash flow; their credit ratings remain high; and a CDS spread in the 50-to-75 basis-point range is a pricing adjustment, not a distress signal. From this perspective, the widening spreads are the market doing its job — attaching a modest risk premium to a genuine increase in leverage — and the equity selloff is noise layered on top of a funding round that succeeded on its own terms. If AI revenue grows into the infrastructure, today's spreads will look like a rounding error.
That argument is correct as far as it goes, but it answers a different question. Nobody is predicting default at these companies. The repricing thesis is not about solvency; it is about the cost of capital and the margin of safety. Even a successful debt raise reprices the risk: the next tranche will be priced off wider spreads, the next project must clear a higher hurdle rate, and the equity multiple that justified the buildout at 2% money does not survive unchanged at 5%. The counter-thesis also assumes revenue catches up on schedule — and the OpenAI revenue gap is the first public data point suggesting the catch-up may lag the spending.
The specific signal that would falsify the structural-repricing view is straightforward: if five-year CDS spreads on the hyperscalers fall back below 40 basis points and stay there for three consecutive months while issuance remains above $100 billion per quarter, the market will have demonstrated that supply can be absorbed without a durable repricing, and the cyclical view wins. Until then, the burden of proof sits with the dip-buyers.
What Comes Next: Beneficiaries, the Exposed, and the Watch List
Short term, the pressure is on the most leverage-dependent corners of the trade: data-center SPVs, neo-clouds, and equipment suppliers whose valuations embed aggressive AI-revenue assumptions. Convertible bonds and high-yield tech debt, which Barclays flagged as widening materially, are the most sensitive to a further supply shock. The 10-year Treasury yield — which fell more than six basis points to around 5.23% on Thursday after approaching 5.36% in the morning — remains the master variable: a move back toward 5.37% would likely extend the equity weakness, while a sustained break below 5% would ease the discount-rate pressure.
Medium term, the companies with the strongest balance sheets and the most certain AI revenue — the distribution toll booths Boloor named — are best positioned to absorb wider funding costs and potentially gain share as weaker borrowers pull back. The exposed are the borrowers whose AI economics depend on cheap, endless capital: any company financing a multi-year buildout at the top of the rate cycle with revenue that has not yet materialized faces a narrowing margin of safety with each repricing.
Long term, the structural question resolves on one metric: does AI revenue grow fast enough to service the debt raised to build the infrastructure? The scenarios split cleanly. In the base case, issuance continues at an elevated pace, spreads settle modestly wider than 2024 levels, and the market digests the supply over several quarters with intermittent volatility. In the upside case, AI revenue accelerates, leverage ratios fall as cash flow catches up, and today's CDS levels become the bottom of a long bull market in tech credit. In the downside case, revenue disappointments multiply, CDS spreads push toward triple digits, and the funding model that powered 2025 and 2026 forces a genuine retrenchment in AI capital spending.
The watch list is concrete: quarterly hyperscaler issuance volumes against the $100 billion-per-quarter threshold; five-year CDS on Oracle, Microsoft, Alphabet and Amazon against the 40-basis-point falsification line; and the 10-year Treasury yield against the 5%–5.37% range. The first of these to break will tell investors whether they are living through a cyclical dip or the beginning of a structural reset in the cost of funding the AI age.
The market is not pricing an AI bust; it is pricing the end of free money for the AI buildout — and those are two very different risks.
Explore more exclusive insights at nextfin.ai.
