Credit risk associated with AI hyperscale data center operators is now breaking beyond the technology sector and spreading into wider markets.
Brian Garrett, a top derivatives trader at Goldman Sachs, issued a warning in his latest weekend preparation report. He noted that stress in the AI-related bond market—including widening credit spreads for hyperscalers, a significant expansion in credit default swaps (CDS), and increased concessions for new debt issuance—is now transmitting to the broader market. Concurrently, the S&P 500 index is increasingly failing to accurately reflect the actual performance of ordinary stocks, with divergences between the index and individual stocks becoming concentrated across asset classes.
Signals of internal market stress have clearly intensified. The Goldman Sachs "panic index" surged by 5.5 points this week alone, a magnitude typically seen only during periods of extreme market stress. Garrett stated that the actual feel on the trading desk last Friday was far worse than what the VIX reading of 18 suggested. He pointed to emerging signs of "partial capitulation selling" in the tech sector but added that it remains necessary to "observe subsequent developments."
AI Capital Expenditure Concerns Spread to Credit Markets
Garrett's core assessment is that the capital expenditure wave from AI hyperscale data center operators has firmly established itself as a primary driver of the global credit impulse. Should this logic develop cracks, the impact would not be confined to equity markets.
Over the past several weeks, market skepticism regarding hyperscalers' return on investment has resurfaced, directly transmitting to credit markets. This has led to a persistent widening of related credit spreads and put pressure on semiconductor and memory chip stocks. Garrett noted that Alphabet, Google's parent company, and Tesla are both scheduled to report earnings this Wednesday. Both are seen as bellwether cases for "return on capital expenditure," and the market will use these reports to re-test the validity of the AI investment thesis.
Garrett's key observation this week was pointed: "If you performed well in the first half of 2026, you are likely having an extremely tough time in July."
Technology Sector Faces Historic Selling Pressure
Data from Goldman Sachs' prime brokerage shows selling in the technology sector has reached record levels. Hedge funds have been net sellers of US information technology stocks in 6 out of the past 8 weeks. The cumulative selling magnitude is the largest in Goldman's records spanning over a decade, comparable to the selling wave in the summer of 2024.
The proportion of total exposure and net exposure to the Information Technology sector within the US prime brokerage book peaked at 5-year highs of 23.4% and 26.3%, respectively, in early June. However, in just about six weeks, these figures have retreated to 19.4% and 14.7%, placing their percentile rankings over the past year at the 32nd and 2nd percentiles, respectively.
Garrett specifically highlighted: "The persistent and large-scale selling since early June indicates technology investors are significantly reducing long positions, with signs of partial capitulation beginning to appear." The TMT sector was the worst-performing and most-sold US sector this week.
Surge in Panic Index and Accelerating Market Distortions
Signals from the derivatives market suggest internal market stress exceeds what surface-level data indicates. The Goldman Sachs panic index surged this week from lows around 1 to highs above 6, a weekly increase of 5.5 points—a magnitude typically associated with extreme market stress events.
Meanwhile, individual stock volatility has significantly amplified, but trading volume has not expanded in tandem. Garrett believes this indicates investors still lack confidence to make substantial portfolio adjustments. Last Friday, DRAM saw a daily swing of 13% but closed flat, while the Philadelphia Semiconductor Index swung 7% but closed down 1%. Garrett noted that the average market capitalization of SOX component stocks is close to $500 billion, and such extreme volatility for companies of this size is itself an anomalous signal.
Furthermore, Garrett observed a puzzling phenomenon: there were large-scale Market-on-Close (MOC) imbalances at the close that moved in the opposite direction of the S&P 500's intraday trend. He attributed this to the prevalence of leveraged products, inverse ETFs, and zero-days-to-expiration (0DTE) options. He believes the current intraday technical landscape for the S&P 500 is "fundamentally different from the past."
Momentum Strategy Nears End, But Risk Not Fully Cleared
The Goldman Sachs High Beta Momentum Long-Short basket has declined 32% from its peak. Garrett's team believes the unwinding of momentum strategies is entering a "late-stage phase" for three reasons: first, the magnitude of decline is similar to comparable historical events (46% drop in 2020, 45% drop in 2021); second, positioning has clearly cleaned up, with momentum long exposure sharply contracting; third, there is currently a lack of clear fundamental catalysts, with AI capital expenditure pressure being the main drag.
Regarding implied correlation, despite a slightly deteriorating macro backdrop, current implied correlation remains near 20-year lows. Garrett argues this means the cost of hedging against the real pain points in individual stocks is significantly higher than hedging the S&P 500 index. The latter, however, is increasingly failing to represent the performance of actual investment portfolios.
Comments