The US corporate credit market is undergoing a profound stress test triggered by the wave of AI capital expenditure, and Goldman Sachs' own credit strategists have publicly acknowledged that their previously set pessimistic scenario may not be pessimistic enough.
Last week, Meta earnings triggered a sharp widening of AI credit spreads, with its credit default swaps (CDS) hitting a record high, and Goldman Sachs' AI basket falling 23% cumulatively since June. Tony Pasquariello, head of Goldman Sachs' hedge fund product division, stated bluntly in an internal weekly report that in his multiple conversations with credit market veterans, the term "melting down" was repeatedly mentioned.
Meanwhile, Goldman Sachs' chief credit strategist Amanda Lynam acknowledged in a latest report that US investment grade (IG) and high yield (HY) credit spreads are rapidly approaching the previously set peak of the downturn scenario, and she expects them to "modestly exceed" that range, with risks "clearly skewed toward further widening."
This turmoil has spread from the corporate bond market to the sovereign bond market and is beginning to extend into European sovereign debt territory. The spread between French 10-year government bonds and German bunds has widened to 178 basis points, the widest level since the 2011 European debt crisis, with the French 10-year yield breaking above 5% for the first time since 2002. The market is facing a rare combination of AI financing pressure, year-end deleveraging sentiment, and European sovereign risk.
The AI bubble spreads from corporate bonds to the broader bond market
For most of the past year, the debate surrounding AI focused primarily on equity valuations鈥攑rice-to-earnings multiples, capital expenditure guidance, and whether Nvidia's customers are also its investors. This debate has now formally evolved into a credit market issue.
Meta earnings became the trigger. The company reported its first negative free cash flow quarter since its 2012 IPO, and its CDS immediately hit a record high, with the stock price falling sharply. Brad Shelofsky, head of Goldman Sachs' investment grade credit trading, characterized last week as "a major repricing of AI risk, and a comprehensive, aggressive deleveraging of crowded positions."
In his weekly report, Pasquariello identified the core logic behind the AI trade evolving into a bond market problem: there is no issue on the demand side, and large tech companies "continue to execute well," but the problem lies in everything needed to meet that demand鈥"as forward spending commitments continue to accumulate, the market is asking very pointed questions about the supply of necessary raw materials, with the two most prominent variables being power and capital."
Goldman Sachs rates trader Mitchell Cornell described the transmission chain: a market regime of rising rates and rising volatility is partly driven by "AI capital expenditure issuance," which subsequently "shakes the confidence of some highly leveraged AI companies." Dallas Fed President Lorie Logan said she estimates the policy rate target range needs to be raised by at least another 50 basis points. The AI capital expenditure business model built during the era of cheap money is now facing a fundamental challenge from the monetary environment.
Goldman Sachs' pessimistic scenario has been overtaken by reality
Goldman Sachs credit strategist Amanda Lynam had previously set the downside scenario peaks for the US credit market at: investment grade spreads of 93 basis points and high yield spreads of 335 basis points鈥攖his was her "bear market baseline." However, in the latest report, Goldman Sachs admits this preset is no longer pessimistic enough:
"For the dollar market, we previously used the March 2026 spread peak as the baseline for our downside scenario (IG 93 basis points, HY 335 basis points). However, we are rapidly approaching these levels and now expect spreads to modestly exceed the above range."
Lynam listed three pressures currently facing the credit market: first, considerations of rising capital costs; second, the multi-year financing wave brought by AI-related investment expectations; and third, positioning and year-to-date performance dynamics, which "may be exacerbating the recent 'risk-off' move." She believes that from now until early December, risks are "clearly skewed toward further widening."
Nevertheless, Lynam still maintains that "a substantial and sustained widening of credit spreads would require a sharp downturn in economic activity, which is not our baseline scenario," and she maintains her year-end target of spreads "modestly tightening" from current levels鈥攃iting the seasonal tightening effect typically brought by slower December supply. However, hyperscaler tech companies like SpaceX and Meta are preparing the largest investment grade bond deals of the year, and this optimistic assumption faces a severe test.
Hyperscaler financing pressure ignites the market
The two core sources of market pressure both point to financing rather than revenue.
Regarding SpaceX, Musk stated on the earnings call that there is a "small but non-zero" possibility that more than 20GW of superintelligent computing power will come online next year, and he expressed high confidence in deploying approximately 10GW of incremental computing power. However, SpaceX had only 1.7GW of computing power as of the end of the third quarter and plans to reach 3GW by year-end. The market is "highly focused on the cost of approximately $50 billion per GW"鈥攁t 10GW, the capital expenditure scale for a single company would reach $500 billion.
Subsequently, SpaceX's previously reported $40 billion chip financing plan (originally scheduled for 2027) was disclosed as "possibly pricing next week." Affected by this, the SpaceX bond curve flattened significantly: the SPCX 5.35% 2031 bond spread widened by 55 basis points during the week to T+225, and the 6.65% 2056 bond widened by approximately 25 basis points to T+285.
Meta, meanwhile, left a rare confession in its 14C filing, pointing directly at the narrowing capital market window: "The special committee also recognizes that timing is critical to lock in capital before competitors exhaust various financing channels. Other prominent companies have recently completed large-scale financings, and more are expected to follow in support of the same AI growth objectives鈥攖his has led to an increasingly challenging financing environment."
Nvidia CEO Jensen Huang put it more bluntly: "The demand for capital far exceeds the supply of capital. This problem will eventually be resolved." But as Shelofsky pointed out, the way it gets resolved will be through price鈥攎eaning higher spreads.
The bond market suffers its "worst week in history"
Looking at the specific data, the market damage is striking.
The Goldman Sachs investment grade hyperscaler basket widened by 17 basis points during the week, "the worst single-week performance since July"; the 30-year hyperscaler basket is 85 basis points wider than the IG100 (excluding hyperscalers), having widened by 90 basis points cumulatively since the basket was established in February. In other words, the market is charging nearly junk-grade premiums to the world's largest and wealthiest tech companies to finance their 30-year data centers.
The damage at the positioning level is equally severe. According to Goldman Sachs prime brokerage data, Thursday saw the largest long selling in over a year, with hedge funds in a deleveraging state for seven of the past eight trading days; the information technology sector experienced its largest net selling since September 2024. The Goldman Sachs hedge fund VIP pair basket posted its worst single-week performance in 15 years, falling 4.1% over five trading days; the high momentum basket gave back all of its year-to-date gains, with a drawdown of 43%. The Goldman Sachs TMT AI vs. AI risk basket fell 10% in a single day on Thursday, the worst single-day performance since the DeepSeek shock; the basket was up 94% year-to-date in June and has now turned to a 13% decline for the year.
The Goldman Sachs EMEA distressed debt team described deeper market fissures in its weekly report: CCC-rated spreads widened by 55 to 65 basis points during the week to 1,320 to 1,340 basis points, down 3.5% month-to-date; global investment grade spreads widened to a 17-month high of 129 basis points; only 10.1% of the loan market is trading above par, with 74.1% between 90 and 100, "leaving extremely limited buffer against further spread widening"; the primary market is "essentially closed, with opportunistic issuance shelved, deals delayed, and some marginal trades withdrawn."
Regarding the continued decline of Virgin Media bonds, the team stated directly: "We have received numerous inquiries about the reasons for the decline, but at this point, we can no longer provide an explanation beyond technical factors."
European sovereign risk adds fuel to the fire
Just as AI financing pressure has yet to subside, the European sovereign debt market has once again become a new source of pressure.
The spread between French 10-year government bonds and German bunds has widened to 178 basis points, the widest level since the 2011 European debt crisis; the French 10-year yield has broken above 5% for the first time since 2002. French bank subordinated debt widened by approximately 25 to 30 basis points during the week, and the European high yield credit index Xover widened by 23.875 basis points.
The Goldman Sachs EU macro credit team noted a concerning signal: credit performance is beginning to underperform rates, "feeling like the first time in a long while." More worryingly, in the process of spread widening, sellers who would typically sell primary protection "simply did not show up" this time, meaning the market is evolving toward a "dispersion-up" regime鈥攏obody is willing to underwrite risk.
Bull vs. bear: Can a year-end reversal materialize?
Faced with this situation, bullish and bearish divisions within Goldman Sachs persist, but the language on both sides is trending cautious.
Pasquariello acknowledged that last week the market "really hit the bone," with Thursday's tech exposure selling being the largest in the past two years, and said he is "licking his wounds" and that "swimming near the shore is perfectly fine right now." But he also noted that Goldman Sachs still expects 18% S&P 500 earnings per share growth in 2027, and "with double-digit earnings growth, bear markets are extremely rare."
Shelofsky admitted he "fell into the mean-reversion trap too early," having originally expected Meta's bond issuance to create room for a fourth-quarter rally, but financing pressure concerns instead intensified last week. He still expects a short squeeze similar to July at some point in the fourth quarter, but only if Meta and SpaceX's chip deals are completed smoothly, and he warned that year-end sentiment poses an additional headwind: "Compared to July, the trickiest part of this round is that the market broadly wants to deleverage amid year-end performance pressure... Participants are all looking to reduce size, protect full-year performance, and close out the year defensively."
Next week, the market will see $3 billion to $3.5 billion in investment grade primary supply, as well as earnings from Google, Amazon, and Microsoft and Oracle's investor day, with "an extremely wide range of outcomes."
As Goldman Sachs' own data presents: credit strategists say spreads will exceed the downside scenario; the investment grade trading desk says hyperscaler bonds just experienced their worst week since July, with 30-year bonds 85 basis points wider than the rest of investment grade; the distressed debt team sees CCC spreads approaching 1,330 basis points, the primary market essentially closed, and hedge funds suffering their largest single-day drawdown since the pandemic; French-German spreads are the widest since 2011; and yet the year-end spread target remains tighter than today鈥攂ecause of "December."