The US stock market is standing at an unsettling peak: the index is at an all-time high, but that strength is being powered by a single theme.

Tony Pasquariello, who heads hedge fund coverage at Goldman Sachs, acknowledged in his latest client note that the current US equity rally is "basically just one big trade" 鈥?artificial intelligence. He maintained his overall bullish stance on US stocks, but his language has clearly become more cautious, and he is no longer willing to assert that the next 5% move will be upward. At the same time, Goldman Sachs data show that about one-sixth of S&P 500 constituents have fallen more than 50% from their highs, while the index itself is only a step away from a record high.

This split structure is fueling deeper concerns about concentration risk. Rates keep climbing, market breadth is deteriorating, and positioning is extremely crowded in the AI sector. Multiple stress signals are flashing at once, leaving bulls defended mainly by whether Nvidia and Micron can deliver on expectations in the coming earnings season.

Extreme Divergence: A Bull Market That Exists Only in Certain Corners

The S&P 500 briefly touched an all-time high of 7,818 this week before pulling back to 7,765 after a revenue warning from OpenAI. On the surface, the bull market flag is still flying, but the internal structure has become highly distorted.

Data compiled by Goldman Sachs colleague Brian Garrett reveals the depth of this divide: over the past three months, about 45% of S&P 500 constituents have been negatively correlated with the index itself, a record high.

Goldman Sachs' Adrien Simonet quantified it further: since August 27, the S&P 500 has risen 1.28% overall, while the "S&P 500 ex-AI" (SPXXAI), which strips out AI-related stocks, has fallen 5.19%. The rolling 30-day gap between the two is close to its largest since the AI trade began in January 2023.

Meanwhile, the Russell 2000 has underperformed the Nasdaq on 16 of the past 20 trading days, lagging by about 9 percentage points. Pasquariello described this market as an "extremely narrow rally" 鈥?US large-cap technology stocks are charging ahead like a locomotive, while rate-sensitive and cyclical sectors remain under pressure.

Positioning Looks "Clean," but Risk Is Concentrated in One Place

Even with the index at a record high, overall market positioning is surprisingly low. Goldman Sachs Prime Book data show that in September, hedge funds' net exposure to US equities was at one-year, three-year, and five-year lows. Net leverage in fundamental long-short strategies fell for a third straight week, dropping to its lowest level since Trump announced reciprocal tariffs last year. Goldman Sachs' sentiment indicator is also hovering near multi-year lows.

Pasquariello's interpretation is that being underweight US equities has itself become a crowded trade, and for institutions worried about missing a move to 8,100 on the S&P 500, the cost of call options is relatively cheap.

However, low net exposure does not mean low risk, because gross exposure remains elevated and the remaining positions are extremely concentrated in the same direction.

In September, technology was the only sector that hedge funds bought on a net basis, with purchases the largest since February 2025.

Goldman Sachs Prime Book data show that net exposure to the "Mag 7" now accounts for about 22% of total US equity exposure, the highest on record since data began in early 2022. Semiconductor positioning stands at 12%, double the level at the start of the year. Simonet added that leveraged semiconductor ETF assets have reached about $115 billion, which would create a mechanical amplification effect in a market downturn.

On Thursday, after the OpenAI news hit, Goldman Sachs' TMT trading desk recorded more than $1 billion in net selling in semiconductors, AI, and large-cap technology names 鈥?a small preview of what could come.

Earnings Growth Looks Strong, but It Relies Heavily on a Handful of Companies

With earnings season about to begin, Pasquariello raised a third point: who wants to short in the face of an expected 28% year-over-year increase in S&P 500 earnings? That figure has been revised up from last week's 27% estimate.

But this, too, is a story that depends on where you look.

Goldman Sachs' Ben Snider wrote in his Q3 earnings preview that earnings growth for the median S&P 500 stock has slowed to 9% in Q3 from 14% in Q2. Broken down by sector, technology growth is as high as 64%, while energy is 122%.

Goldman Sachs' Wilson estimates that 68% of S&P 500 earnings growth this quarter will come from the top 10 contributors, up from 48% last quarter. Micron and Nvidia alone account for half of that 68%. More notably, Goldman Sachs' 27% full-year earnings growth forecast is already a cycle high in its own model, and growth is expected to slow in each of the next three quarters. Capital expenditure growth at hyperscalers, which Snider estimates at 116%, may also peak this quarter. A peak in growth and a peak in capital spending arriving together, combined with stock prices and positioning at historic extremes, is the kind of combination that tends to be described only in hindsight as having been "obvious all along."

Rates Move from Headwind to Systemic Risk

This is the most striking change in Pasquariello's language. Ten days ago, he described rising rates as a "headwind" for stocks but explicitly ruled out anything "disruptive." In his latest note, he upgraded that wording: rising rates are becoming a "greater risk" for sovereign and corporate debt.

Market data confirm the shift. On Thursday, the 30-year Treasury auction cleared at a yield of 5.618%, the highest since August 2000, while the 10-year Treasury yield closed at 5.23%. Goldman Sachs economists expect the "Warsh Fed" to raise rates again in December after hiking to 3.75%-4% in September. Goldman Sachs' Rikin Shah calculates that of the 108-basis-point selloff in 10-year yields, 98 basis points came from real rates, and the 30-year real yield at 3.33% is already near the upper end of its range since 2010. His conclusion is that the Treasury market is searching for one of two thresholds: the level at which AI financing demand becomes sensitive to rates, or the level at which some other part of the economy breaks first.

On the corporate credit side, Goldman Sachs credit strategists noted that dollar investment-grade credit has returned -292 basis points year to date and highlighted software issuers facing refinancing pressure from maturing loans. Goldman Sachs economists estimate that current rate levels could drag on economic growth by 0.5 percentage point in 2027 through housing, consumption, and capital spending channels.

Simonet's warning was especially direct: a term premium shock will not distinguish between good and bad AI stories; it will reprice all 10-year cash flows at once. He sees the real left-tail risk as the Federal Reserve being forced to raise rates more than expected to defend long-end credibility, with CPI data as the near-term trigger.

Both Bull and Bear Cases Point to the Same Bet

In the note, Pasquariello quoted an investor he clearly respects a great deal, and he called it the most important line in the whole report:

"You know, this is basically just one big trade."

He did not argue with it. Looking at six judgments side by side 鈥?trend, positioning, earnings, rates, concentration 鈥?they turn out to be different sides of the same observation: the trend is up because of AI; positioning is low except in AI; earnings growth depends on AI and its energy needs; rates are rising partly because of AI's financing needs. The bull case and the bear case both point to the same target.

Pasquariello also offered a friendlier "technology and energy side by side" perspective: over the past 11 years, a 50/50 portfolio of the two sectors delivered positive returns in 10 of those years, with an average annual return of 18.7% and a Sharpe ratio of 1.2. But that, too, is simply describing the same trade and its energy constraint.

Seasonality and the Midterm Elections: The Final Variables

Pasquariello's sixth judgment was relatively brief: Q4 seasonality is fairly favorable, but he expects the November 3 midterm elections to bring a wave of volatility.

Goldman Sachs' Alec Phillips noted that prediction markets give Democrats a more than 90% chance of retaking the House, 65% for the Senate, and an 8.9 percentage point lead in the generic ballot, with a "sweep" now the market's base case.

Meanwhile, the VIX closed at 15.4 on Thursday, but Goldman Sachs' volatility trading desk reported that buyers purchased about $10 million of vega in year-end S&P 500 puts in roughly six hours that day. Someone is buying insurance against "friendly seasonality."

A Market with Only One Exit

Pasquariello ultimately offered three conclusions: the bull market holds, but it depends on where you look; near-term risk-reward is unclear, so "mind the speed limit"; stick with the two fastest horses 鈥?the United States and Japan; and hedge with a combination of long equities and short rates.

That hedge itself is telling: if the recommendation is to own stocks while shorting bonds, it is effectively saying both will fall together, and "shorting rates" only makes money if yields keep rising 鈥?which is precisely what the 85% of constituents already in deep drawdown can least withstand, and which will ultimately become a financing-cost pressure on the other 15%.

The bull case requires Micron and Nvidia to deliver, hyperscaler capital spending not to peak, the Fed to stop after two hikes, and someone to be willing to buy long bonds at a 5.6% yield. The bear case needs only one of those to fail.

One big trade is wonderful while it is rising. But by definition, it also means the entire market has only one exit.