Hedge fund returns have "evaporated by half" this year, with losses especially severe in recent weeks. Goldman Sachs partner: Wait for "junk-rated" assets to stop falling.

The wave of AI trading reversals continues to impact the market, with hedge fund positions taking heavy losses. However, insiders at Goldman Sachs believe the sell-off has become excessive, and the pendulum may have swung too far.

Goldman Sachs Prime data shows that Asia fundamental long-short funds are down 7% this month, with year-to-date returns plunging from 40% at June’s peak to just 16%, meaning over half a year’s gains have been wiped out in the past four months—most of the losses occurring in recent weeks. Global AI stocks have seen net selling for six consecutive trading days, with the cumulative net flow Z-score over six days dropping below -3, marking the steepest reduction in positions since the launch of ChatGPT. Goldman Sachs analyst Mark Wilson points out that sentiment, pricing, and positioning momentum have all turned sharply negative, and currently, the pendulum has "most likely swung too far in the opposite direction."

Meanwhile, the market's focus is shifting from AI demand to credit risk in capital structures. Cloud computing company CoreWeave saw its share price plummet 15% in a single day, with five-year CDS hitting a record high. Its July low has become the critical technical level to watch. Wilson warns that until the decline of junk-rated AI assets represented by CoreWeave stops, any rebound in AI stocks is likely just a position squeeze, not a fundamental repricing.

Goldman Sachs Prime data outlines the precise extent of this drawdown. TMT long-short funds are down 5% this month, with year-to-date returns narrowing from 24% in June to 12%; global fundamental long-short funds are down 2% this month, with year-to-date returns dropping from 17% to 10%; European long-short funds are flat this month, but still post an 8% gain year-to-date.

Pressure on positions is also apparent. Goldman Sachs Prime books show that global AI stocks (GSXGGLAI) have seen net selling for six consecutive days, with the six-day cumulative net flow Z-score dropping below -3, surpassing the “second fastest sell-off YTD” set the previous week, and coinciding with the sharpest single-day quant strategy drawdown since August.

Wilson specifically mentions a clear trading logic maintained for three and a half years—longing “AI winners” and shorting “AI risk assets”—a strategy which at one point this year returned over 90%, but has nearly halved since late June, with September’s attempted rebound thwarted and now retesting lows.

Wilson attributes this week’s most violent fluctuations to two specific triggers.

First, CoreWeave co-founder McBee said in an interview that life insurance companies are providing “unlimited” capital to AI—“all the money in the world.” This statement triggered a reevaluation of private credit pricing and heightened concerns over the weakest links in the AI funding chain.

Second, debates around the business models and economics of cutting-edge AI models have intensified—including OpenAI CFO Sarah Friar publicly warning enterprises to be cautious with cheap open-source models. Wilson expects this debate to “become clearer and more transparent in the coming weeks,” directly influencing judgments on capex scale, credit quality, and AI value attribution.

Both developments point to the same core question: Within the layered stack of AI capital, who ultimately pays for expensive funding? Goldman Sachs credit strategists identify “emerging cloud service providers and frontier lab-related AI company bonds” as the epicenter of the funding cost debate.

Wilson does not agree with an entirely bearish stance. He acknowledges that the “second derivative” of capex and earnings growth will most likely slow from current levels, but stresses that prices have already reflected much of this expectation.

His core argument comes from a remark by OpenAI CFO Sarah Friar at a Goldman Sachs event about a month ago: “We’ve been criticized in the past two years for over-investing in compute, but this year has been entirely worth it. The computing power I bought a year ago can now be sold in the market for 3-5 times what I paid. The only bad news is that we still don’t have enough—we should’ve bought more.”

Based on this, Wilson reiterates his key view: market leadership is rotating from “AI spend beneficiaries” to the “largest AI investors”—that is, the mega-cap technology firms now delivering investment returns via Q3 earnings. He notes that most of these companies can fund their capex internally, are relatively insulated from value attribution disputes, and have begun to outperform the broader AI winner basket.

Wilson points to a notable change in the bond market during this correction.

Although the drivers for stronger bonds—growth risks and unexpected funding events—aren’t what equity investors typically hope for, the 10-year US Treasury yield failed to rise further after CPI and PPI data, and retreated markedly on Thursday, breaking below a steep uptrend. Wilson believes this "shows a kind of fatigue in the market never seen before."

However, Wilson also warns that energy price pressures have yet to ease similarly, with December WTI futures approaching $100, and interest rate risks remaining significant.

Looking ahead on supply, Goldman Sachs’s Ben Snider estimates that, including expiring lock-ups, net US equity market supply next year will see its largest increase since 2000. Wilson is relatively calm about this, arguing that history shows as long as the market believes equity raising is being used for productive investment, higher supply in itself is no obstacle to index gains—the key is “whether the market believes this.”

Wilson’s bullish case stands on three pillars: positions have been cleaned out, bonds are stabilizing, and compute demand shows no signs of slowing.

However, the key issue has shifted from “is there compute demand” to “who will pay for it.”

CoreWeave’s credit risk trend shows that the lower end of the AI capital stack—junk-rated assets—remains under pressure. Until pricing in this segment stabilizes, any rebound in AI stocks is more likely to be a short squeeze than a true consensus on fundamentals.CoreWeave’s July low is the key technical level to watch next week.