Tech stocks suffered a selloff this week, and Goldman Sachs has issued a fresh warning that the surge in AI-related debt is quietly building pressure in credit markets.

In the 2026 US congressional midterm elections, Democrats won across the board in both the Senate and House as well as multiple gubernatorial races, with many victorious candidates pledging to restrict data center construction and tighten industry regulation. The market is now worried that investment in artificial intelligence infrastructure will be suppressed by policy.

On Thursday (November 5), the S&P 500 plunged 4.4% in a single day, while the Nasdaq 100 tumbled 6.4%, wiping out more than $3 trillion in market value and marking the worst one-day drop for tech stocks since the COVID-19 pandemic.

Yet the stock market shock may only be the opening act. Goldman Sachs credit strategist Amanda Lynam warned in a latest report that AI debt supply is flooding into the market at a pace far faster than anyone expected, and that the risk to credit spreads over the coming weeks is "clearly skewed toward further widening."

The report said that after the tech stock rout, pressure in the credit market is quietly building: as of the close on November 4, US dollar investment-grade credit spreads had narrowed to just 4 basis points from their March high, while high-yield bond spreads were only 11 basis points from their March high, and that was before Thursday's stock market plunge.

In the single week before the election results came out, five major hyperscalers, Alphabet, Amazon (AMZN.US), Microsoft (MSFT.US), Meta (META.US) and Oracle (ORCL.US), together completed five bond issuances with a total scale of as much as $67 billion, pushing their cumulative issuance for the year to $296 billion.

That figure has already exceeded the upper limit of Goldman Sachs' forecast for all of 2026, which was $25 billion. In other words, Goldman Sachs' forecasting model is being shattered by reality once every 48 hours.

$67 billion in one week, full-year forecast scrapped early

The speed at which Goldman Sachs' forecasts have failed is astonishing. According to a report published by Amanda Lynam on Tuesday, as of October 29, the five major hyperscalers had completed three bond deals totaling $41.7 billion, bringing cumulative issuance for the year to $271 billion.

Just two days later, two more deals landed, adding another $25 billion in total scale, equivalent to Goldman Sachs being proven wrong once every 48 hours.

The size of individual deals was equally stunning to the market. Microsoft completed its first US dollar bond issuance since 2017, with a scale of more than $25 billion, immediately overturning Goldman Sachs' earlier judgment that Microsoft would not need to borrow at least this quarter. That judgment became obsolete in less than a day.

Oracle, just five days after announcing its financing plan, exhausted its originally four-year financing "ceiling" in a single morning.

Goldman Sachs offered two explanations for this better-than-expected phenomenon. First, the five major hyperscalers collectively posted negative free cash flow in the third quarter of 2026, with capital expenditure exceeding operating cash inflows. Second, some issuances were "early lock-ins" of financing plans originally scheduled for 2027.

In Goldman Sachs' words, "borrowers know full well that the issuance window may not stay open forever."

Issuance for the year has already reached an amount equivalent to 36% of hyperscalers' 2026 capital expenditure, higher than Goldman Sachs' initial assumption of a 33% debt financing ratio.

Looking ahead to 2027, Goldman Sachs expects direct bond issuance by hyperscalers to reach $440 billion, and combined with about $300 billion in investment-grade and high-yield project financing, the total AI-related debt supply pipeline will approach $740 billion. That does not even include any emerging cloud computing firms, chipmakers or utility companies.

Spread pressure builds, credit market "saturation" alarm sounds

Goldman Sachs made clear that from now until early December, credit spread risk is "clearly skewed toward further widening." In the US dollar market, Goldman Sachs expects spreads to slightly exceed their March highs (93 basis points for investment grade and 335 basis points for high yield). In the euro market, it views April 2025 levels as the reference for a near-term downside scenario (128 basis points for investment grade and 446 basis points for high yield).

Goldman Sachs' year-end targets imply a slight narrowing of spreads from current levels, but Lynam admitted in the report that this mainly reflects the seasonal tightening pattern when supply dries up in December, "and supply technicals are especially critical this year."

In terms of excess returns, US dollar investment-grade bonds have delivered only 0.1% excess return relative to Treasuries this year, high-yield bonds 0.0%, and euro high-yield bonds minus 1.0%. After ten months of record supply, credit investors have gotten almost nothing.

This may also explain why Goldman Sachs recently quietly abandoned its preference for investment-grade bonds over high-yield bonds, shifting instead to favor BBB-rated allocation relative to A-rated and AA-rated.

Pricing pressure on large deals also cannot be ignored. So far this year, the US dollar investment-grade market has seen 27 deals sized at more than $10 billion, totaling $501 billion, about 2.5 times the scale of similar deals last year. Of these, 14 deals, accounting for about 66% of the total amount (roughly $330 billion), are AI-related.

The new issue concession for AI-related deals has averaged about 18 basis points so far in November, far higher than the less than 7 basis points for non-AI deals and significantly above the roughly 8 basis point average for AI deals since early 2025.

In other words, the world's highest-rated borrowers are paying premiums close to junk-bond levels to complete issuance, and each premium further widens the secondary market curve.

Election shock: data center moratoriums and private credit's last line of defense

On the policy front, Goldman Sachs economists believe the election results are unlikely to change the data center and AI regulatory framework at the federal level, as relevant legislation would still need bipartisan Senate support and must clear a veto override threshold.

Near-term risks mainly come from the local level: Democratic victories in state and local elections could trigger data center construction moratoriums in the short term, potentially affecting about 45% of planned US data center capacity, with Texas alone accounting for 25% of that.

Goldman Sachs economists expect these moratoriums to be short-lived and the issue to fade over the longer term. However, credit markets do not intend to wait for the outcome.

Lynam said the high-yield primary market may currently be in a "price discovery phase," with the market paying close attention to the structural design of project financing deals. "Recent price action has fully demonstrated the importance of offtaker quality."

When the syndicated loan market becomes congested, Goldman Sachs views private capital as a backup exit. The private market currently manages $17 trillion in assets (excluding secondaries and fund-of-funds), with $4.6 trillion in dry powder, and has completed more than $225 billion in data center financing since early 2025 (excluding GPU and chip collateralized transactions).

But Goldman Sachs also acknowledged that private credit choosing to become the "lender of last resort" for AI financing in a year when its own default rate has just hit a record high is rather delicate timing.

The race between the monetization clock and the saturation clock

Goldman Sachs' analytical framework ultimately boils down to a race between two variables: accelerated AI monetization means less debt; delayed monetization and growing procurement commitments mean more debt, "depending on market conditions."

The former is at best a 2028 story, while the latter advances every 48 hours: $296 billion in hyperscaler bonds have already landed, total global AI-related debt stands at $687 billion, another $440 billion in direct hyperscaler issuance plus $300 billion in project financing is waiting next year, and trillions of dollars in off-balance-sheet commitments have yet to enter debt statistics.

Goldman Sachs' base case is that the usual December supply slowdown will provide breathing room for its year-end spread targets. But as Lynam said, from now until early December, the balance of risk has clearly tilted toward wider spreads.

The credit market was once the last asset class in the AI trade still giving the benefit of the doubt; with excess returns at zero and issuers openly racing to lock in financing windows, that benefit of the doubt is rapidly being exhausted.

Off-balance-sheet liabilities: trillion-dollar worries beneath the iceberg

What bond issuance data reveals is only the tip of the iceberg. In her Tuesday report, Lynam for the first time systematically quantified hyperscalers' off-balance-sheet obligations: $1.7 trillion in operating and finance lease commitments, of which more than 75% involve leases that have not yet commenced and, under lease accounting standards, are not currently reflected on any party's balance sheet; and $1.9 trillion in other unconditional purchase commitments (covering third-party cloud capacity, servers, networking, power contracts and construction projects), of which $685 billion will come due by the end of 2027 and which exclude contingent backstop obligations.

Goldman Sachs acknowledged that some of these commitments have already been incorporated into consensus capital expenditure expectations, but still treats them as additional cash uses. Every commitment that cannot be covered by operating cash flow will ultimately be converted into bonds, loans or equity financing.

Goldman Sachs put it this way: "Delayed monetization and growing procurement commitments... are key risks driving more debt issuance."

The world's top rating agencies have already begun to sound warnings. Moody's noted on September 22 that hyperscalers' overall financial commitments have grown eightfold since 2023 to $2.8 trillion. S&P warned on October 15 that the AI financing ecosystem's "margin for error is narrowing," and that contingent guarantees and new types of off-balance-sheet financing tools could expand further and put pressure on ratings.

In late October, S&P lowered its rating outlooks on Broadcom and Meta, both citing off-balance-sheet obligations. Goldman Sachs' conclusion is that "evidence of monetization before 2028 is crucial," otherwise ratings will face downward pressure.

However, Goldman Sachs also noted that AI-related companies are generally at the higher end of investment-grade ratings, and even after multiple-notch downgrades they are unlikely to be cut off from market financing channels, drawing an analogy to the slow downward drift in pharmaceutical industry ratings a decade ago.

It is worth noting that Meta explicitly stated in its SEC filing that the company "recognizes that timing is critical and that it needs to lock in capital before competitors exhaust various financing channels."

When issuers themselves are publicly racing to tap the bond market's balance sheet, that balance sheet itself becomes a scarce resource.