While the S&P 500 has posted gains for nine straight weeks, structural risks are quietly building beneath the surface.

On June 2, Lee Coppersmith, a top trader in Goldman Sachs's sales and trading division, issued a warning in a new report: while the index-level rally appears smooth, the underlying dynamics are becoming "increasingly unsettling." He noted that market bets on AI have evolved from being fundamentally driven to a self-reinforcing cycle amplified by market structure itself—positions are more crowded, leverage is higher, concentration is greater, yet the cost investors pay to protect against downside risk has fallen to historic lows.

The most telling indicator of this contradiction is single-stock skew: the average one-month put/call option skew for S&P 500 constituents has fallen to the lowest level in Goldman Sachs's dataset history, meaning investors are abandoning downside protection en masse in favor of chasing upside exposure—all while the realized volatility of individual stocks continues to climb.

Goldman Sachs's US Vol Panic Index has also retreated to a near two-year low.

Concurrently, Goldman Sachs Prime data shows that hedge fund gross leverage rose another 2.1 percentage points this week to approximately 323%, hitting a five-year high.

Index Calm Masks Extreme Factor Volatility, Creating a Stark Divergence

Since ChatGPT was launched on November 30, 2022, the S&P 500 has gained roughly 85%, weathering events including the March 2023 regional banking crisis, the August 2024 Japan rate hike shock, the April 2025 so-called "reciprocal tariffs" turbulence, and this year's US-Iran conflict, each time recovering and reaching new highs.

However, Coppersmith's report emphasizes that the primary concern now is no longer the gains themselves, but the growing divergence between index-level calm and the intense rotations occurring at the factor, positioning, and single-stock levels.

The Nasdaq 100 Index rose about 10% in May alone, marking its first back-to-back months of double-digit gains since 2009 and a 32% rebound from its March low. Behind this impressive figure, Goldman Sachs's TMT Momentum Pairs Index has experienced over 25 single-day swings of ±5% this year, compared to just six for all of last year.

The software sector rose about 8% again last week, driven by favorable positioning, better-than-expected earnings, and renewed optimism about the "rate of change" in AI monetization. Yet even within the sector, divergence is evident—Goldman's trading desk consistently observes a clear separation between perceived infrastructure winners in data infrastructure and cybersecurity, and the more challenging SaaS models.

AI Positioning Self-Reinforces, Doubling Leveraged ETF Assets Amplify Market Fragility

Coppersmith specifically highlighted a developing structural risk: the AI trade is no longer driven purely by fundamentals; market structure itself is creating a reflexive feedback loop.

Global leveraged/inverse single-stock ETF assets under management have surpassed $60 billion, having doubled since just early April this year. Products linked to SK Hynix, Samsung, and broader memory exposures have seen explosive inflows as investors persistently chase concentrated exposure to the AI build-out.

These products are structurally short gamma, meaning sustained inflows reinforce momentum, amplify short squeezes, and exacerbate violent factor rotations.

Meanwhile, hedge fund positioning data corroborates this crowdedness. Goldman Sachs Prime data shows hedge funds were net buyers of non-essential consumer stocks for a fourth consecutive week across all major regions, despite the sector's allocation remaining near five-year lows. The technology sector experienced its largest deleveraging week in over a month, yet overall allocation remains near five-year highs.

Downside Hedging Costs at Record Lows, "Reverse-Dispersion" Hedges Regain Attention

The correlation regime implied by current market pricing is extremely benign, which is precisely what concerns Coppersmith most.

S&P 500 implied correlation is near historic lows, with index volatility significantly lower than constituent stock volatility. Historically, this setup has provided a rationale for "reverse-dispersion" hedging strategies—shorting single-stock volatility while going long index volatility.

This strategy works when the market shifts from idiosyncratic dispersion to macro panic, causing correlations to spike suddenly, as seen during the Global Financial Crisis, the COVID-19 shock, the 2018 "Volmageddon," the August 5, 2024, Japan market stress event, and last April's so-called "Reciprocal Tariff Day."

However, Coppersmith acknowledges that holding this hedge has been extremely difficult in recent years, precisely because the opposite dynamic has dominated: narrowing leadership, persistently declining correlations, and volatile single stocks alongside a relatively orderly index.

His summary is blunt: the market is growing more convinced, more concentrated, more leveraged on AI, and increasingly willing to pay for upside convexity rather than buying downside protection. Macro-level panic has subsided, but stock-specific risk has not disappeared—in fact, single-stock skew pricing suggests the opposite.

Earnings Growth Provides Fundamental Support, But Uncertainty Remains

Nevertheless, Coppersmith concedes that this rally is not purely valuation expansion or blind speculation; earnings growth has indeed driven most of the gains.

Prior to ChatGPT's launch, Goldman Sachs's US strategy team forecasted 2023 S&P 500 EPS around $224. Today, the team forecasts 2026 EPS of $340, approximately 52% higher than pre-AI era forward expectations. Since November 2022, about two-thirds of the S&P 500's gains can be explained by earnings growth.

Within the AI supply chain, Goldman's technology research team recently estimated that Agentic AI could increase token consumption by roughly 24 times current levels, thereby creating massive incremental demand for GPUs, CPUs, power systems, cooling infrastructure, network architecture, and overall data center redesigns.

Coppersmith points out that the market still does not fully understand where AI value will ultimately accrue in the supply chain, and this uncertainty itself may be helping sustain the breadth of global AI capital expenditure enthusiasm.

He also notes in the report that historically, major technological shifts have tended to increase concentration rather than dispersion, with the firms best positioned to deploy intangible capital—data, compute power, distribution channels, software ecosystems, and network effects—at scale often capturing a greater share of economic value.