Last week, one word kept circulating through the U.S. credit market: "melting down." Goldman Sachs' head of hedge-fund coverage said in his internal weekly that, after several rounds of conversations with senior credit-market players, that word came up again and again. Goldman's own chief credit strategist admitted in her report that the bearish scenario she had set earlier was "perhaps not yet fully priced."
But look closely at this panic and you notice it differs from past credit crises in one way: the protagonists are precisely the best credits in the market.
Start with rates, the anchor. Public data shows the 10-year Treasury yield briefly topped 5.3% in early October, its highest since 2002; by October 22-23 it had eased back to 4.95%-5.00%, a net decline of roughly 30 basis points over the month. The 2-year sits near 4.7%, the 30-year near 5.3%, the curve is a normal upward slope, and the 10y-2y spread is about 30bp — hardly an inversion. In other words, the rate side has been easing this month, not tightening.
What is actually tightening is credit spreads. The option-adjusted spread on U.S. high-yield (HY) closed at 340bp on October 22. The "downside scenario" peak that Goldman's chief credit strategist Amanda Lynam had set for herself was exactly 335bp — reality has already burst through that script.
1. Goldman's bear-case script, realized ahead of schedule
Goldman's "bear-case baseline" set the investment-grade (IG) spread at 93bp and high-yield (HY) at 335bp. By late October, HY at 340bp had already exceeded it, and IG at 88bp (October 21) was closing in on the 93bp threshold. Goldman admitted in its report: "We now expect spreads to modestly exceed the above range," with risks "clearly skewed toward further widening."
She listed three pressures: higher cost of capital, a multi-year funding wave from AI investment, and positioning plus year-to-date performance dynamics — the latter two amplifying the year-end "risk aversion." So in her view, from now until early December, the direction of spreads is further widening, not tightening.
But there is a percentile point that is easy to miss. Goldman also said that the IG index spread, viewed over a 20-year cycle, still sits around the 10th percentile — still tight in absolute terms. That is exactly what makes this move awkward: panic-driven widening on one side, and an absolute level that is historically not cheap, even rich, on the other. That scissors gap is more worth pondering than "crash or no crash."
2. The protagonists are the best credits
Now the trigger. After Meta's earnings last week, AI-linked credit spreads widened sharply, and Meta's five-year credit default swaps (CDS) hit 119.3bp, a record since listing. Goldman's own AI basket has fallen 23% since June.
There is a striking irony here: the record-high CDS was triggered by Meta, rated AA-/Aa3 — not by any junk-rated company. Oracle is even more extreme — S&P cut it to BBB- in July, the last rung of investment grade, just one notch from junk. Public reports say the roughly $18 billion in loans tied to Oracle's "Project Jupiter" data-center campus in New Mexico have come under pressure; the company issued a force majeure notice, and those loans were quoted at around 85 cents on the dollar in private markets.
So what is really changing in this round of tech debt is another variable: supply. Beyond the surface narrative of "deteriorating credit quality" — none of the big four (Microsoft, Apple, Google, Amazon) has been downgraded, and Moody's has explicitly called their balance sheets "the strongest in the world" — the more important point is that these strongest credits are collectively becoming the biggest borrowers.
3. From $35 billion to $132 billion, the strongest credits became the biggest borrowers
Public data draws this line clearly. Vanguard's calculations show that the five — Alphabet, Amazon, Meta, Microsoft and Oracle — issued roughly $35 billion a year combined on average from 2020 to 2024; that jumped to $93 billion in 2025; and through the first seven months of 2026 it has already hit $132 billion. In between was a roughly $53 billion multi-tranche deal and a rare "century" bond.
Widen the lens a bit and global AI-linked debt financing totals about $466 billion so far this year, versus only $101 billion in the same period last year. The IMF's tally is even more direct: these five hyperscalers alone account for nearly 10% of global investment-grade corporate issuance in 2026.
And the supply wave is far from over. S&P Global expects the top six U.S. hyperscalers (Amazon, Microsoft, Alphabet, Oracle, Meta, plus SpaceX) to spend over $7 trillion on data centers and AI capex between 2025 and 2030, with free cash flow negative at least through next year. In issuance terms: SpaceX is in talks for about $40 billion of financing to buy Nvidia chips, Wall Street is assembling a $60 billion package for Broadcom and Anthropic, and Meta is preparing a new share class, with the market reporting it may pair $30 billion of investment-grade debt with $10 billion of bank credit.
The only pause was September — global AI debt issuance plunged from June's peak of $113 billion to $23 billion. But that looks more like a "gear change" in supply than a "stall," because the biggest deals of the year, SpaceX and Meta, are still ahead.
4. The anchor for pricing spreads shifted from default to supply
This spread-widening has even spilled into sovereign debt: the French 10-year spread over Germany widened to 178bp, the widest since the 2011 euro-zone crisis, and the French 10-year yield briefly topped 5% for the first time since 2002. Goldman described the move as a triple overlay of AI financing pressure, year-end deleveraging and European sovereign risk.
Pull these lines together and the conclusion is not complicated: in this round of tech-bond spread widening, the pricing anchor shifted from "default risk" to "supply shock." Historically, widening spreads meant default expectations were rising; this time ratings haven't moved and defaults haven't picked up, yet spreads are widening because the slope of supply has changed — five companies tripled their annual issuance from $35 billion to the trillion-scale within three years, and the market's capacity to absorb it is finite. Add AI capex pushing free cash flow negative, and even with ratings unchanged the market demands a higher "uncertainty premium." This is not the 2011 euro-zone crisis, when sovereign credit genuinely collapsed; the spread into French debt this time is more a contagion of the same deleveraging and supply sentiment, with no comparable deterioration in credit fundamentals.
On positioning, my judgment is threefold. First, rates actually eased in October, but the market is still pricing roughly a 70% probability of a Fed hike in December — the short end has not loosened, so duration should not be aggressively extended, and the long-end 30-year warrants caution on supply and volatility. Second, spreads are "clearly skewed toward widening" into early December, and the IG absolute level at the ~10th percentile of 20 years is not cheap — so before the supply flood lands, rather than rushing to buy the dip in spreads, it is better to keep your seat at the table and wait for the December window of seasonal supply slowdown before adding. Third, distinguish among issuers: the AAA/AA names — Microsoft, Apple, Google, Amazon, Meta — still have a credit cushion, and their risk is valuation volatility rather than default; the real minefield is names like Oracle, BBB- with negative free cash flow, which, if cut another notch into junk, would trigger index exclusion and forced selling — that is the event risk to avoid in advance.
As a fixed-income researcher holding a U.S. tech-bond portfolio, my read of this move is this: beyond the market's talk of a "bursting bubble," watch how it is repricing "the debt-financing of capex." The strongest balance sheet, once it starts borrowing for the most aggressive capital spending, also gets re-examined by the market. That logic is one Chinese tech companies will sooner or later hold up to their own mirror.
⚠️ Disclaimer: This article reflects personal views and does not constitute investment advice.