Goldman Sachs prime brokerage data shows that hedge funds are fleeing technology stocks at the fastest pace in at least a decade.
Over the five trading days through October 15, the pace at which hedge funds cut their technology holdings set the largest record since Goldman began compiling data in 2016; total trading activity also fell at the fastest rate in nearly three months during the same period.
The reduction came as global technology stocks suffered a sell-off. As AI trades were unwound worldwide, the Nasdaq 100 fell 5.5% last week, its worst performance since April 2025; over the five trading days covered by Goldman's data, the technology sector was also the worst performer in the MSCI World Index.
Goldman data shows that global equities overall saw net selling at a scale reaching two standard deviations of the past year's fund flows. Among them, the information technology sector led declines by a wide margin, with the largest total position reduction in two months; the semiconductor sector suffered the largest dollar-denominated losses.
Still, hedge funds continue to hold massive technology positions. According to a Goldman report, net technology allocation accounts for 23.5% of its global prime brokerage book, still at the 90th percentile of the past five years.
But fund sentiment has turned cautious, with their net leverage ratio — a measure of long positions exceeding short positions — dropping to 48.6%, a one-year low.
The Real Trigger: Credit Cracks in the AI Financing Chain, Not Valuations Themselves
The AI trade has been unwound globally, and the deeper catalyst lies in the financing structure of AI infrastructure now exposing credit risk. Using CoreWeave as a sample: its 2032 bond spread broke through 1,000 basis points, and its five-year CDS hit a record high of 1,117 basis points, implying roughly a 63% probability of default within five years; its stock plunged 15% in a single day, and its bond prices fell below 80 cents on the dollar. Nscale's IPO also faced pricing pressure, with the valuation some investors advocated representing a 37%–47% discount to the $30 billion target.
This points to a thornier problem than "valuations are too high": a structural mismatch between the pace of AI infrastructure spending and the pace of revenue. Expensive outlays on GPUs, data centers, and power occur first, while the revenue side must wait for grid connection, acceptance testing, and actual customer activation; the higher the financing cost, the weaker a project's ability to withstand delays and underutilization.
When the core of market pricing shifts from "how big is compute demand" to "who will pay for the high cost of capital," the first to be sold off will inevitably be the links with the most fragile balance sheets and the greatest sensitivity to financing conditions — semiconductors, new cloud service providers, and highly leveraged AI names — rather than mega-cap technology companies with abundant cash flow.
The Nature of the Divide: How Far Has Deleveraging Gone?
The core disagreement on Wall Street today is not whether long-term AI demand exists, but whether deleveraging has already run most of its course. On the more cautious side (parts of Goldman's team): from a one-year perspective, AI positioning appears to have been relatively cleared out; but shifting to a three-year perspective, global AI stock net allocation remains at the 83rd percentile, and the AI portfolio long-short ratio of 3.27 is still at the 64th percentile. JPMorgan's TPM positioning indicator sits in a neutral range around the 46th percentile, far above the 19th percentile low at the end of July, meaning there is still room for "further deleveraging." Trading feedback from Goldman's Asia team shows that the degree of "capitulation" in this sell-off in the Taiwan market was less than in July, suggesting there may still be selling pressure ahead.
On the more positive side (Morgan Stanley, Bank of America): Morgan Stanley believes the months-long valuation adjustment across most stocks is nearing its end, and after significant declines, technology giants' valuations are close to historical cycle lows, while fundamentals such as prices, orders, and earnings trends for core AI infrastructure chips have not deteriorated to a corresponding degree, with dip-buying opportunities forming after the pullback. Bank of America's bullish logic is similar — the market is already discounting future financing pressures in advance, but spot price resilience and inference demand can still support earnings expansion for a considerable period.
The disagreement between the two sides can be reduced to one judgment: whether the credit contraction on the financing side represents a substantive blow to the long-term AI narrative, or simply a forced cleanup of excessive leverage, while the demand side itself has not deteriorated. Goldman also offers a similar balanced view internally — its analyst Mark Wilson acknowledges that the "second derivative" of capital expenditure and earnings growth will slow from current levels, but emphasizes that prices have already reflected this expectation to a considerable extent, and cites OpenAI's CFO as saying there is no evidence of slowing on the compute demand side.
In short, the current retreat is not smart money systematically abandoning technology stocks, but rather, under the constraint of tightening financing conditions, being forced to compress positions from "high-leverage AI exposure" into "low-leverage certainty holdings." The intensity of the selling reflects the fragility of leverage, not a final verdict on the AI narrative.
Therefore, a rebound in AI stocks requires watching for signs of stabilization in junk-rated AI assets and whether the net leverage ratio stabilizes and rebounds. If the one-year low of 48.6% continues to decline, it indicates deleveraging is still underway; if consecutive stabilization appears in subsequent data, it points to a marginal attenuation of selling pressure. In addition, whether the divergence between the bond market and the stock market converges is also a signal to watch. In this round of declines, the 10-year U.S. Treasury yield actually fell back to a multi-week low even as inflation exceeded expectations, with bonds completing tightening on behalf of the Fed while also pricing growth risk. If credit spreads in the bond market stop widening and junk-rated AI bond yields peak and retreat, that will be a more reliable turning signal than a stock price rebound.
Comments