"Agent vs US Treasury" — Who Will Dominate the US Stock Market?
The wave of AI Agents and US Treasury yields are splitting the US stock market into two worlds: Meta's release of the Muse model boosted its market value by $220 billion in a single week, propelling the Nasdaq's standout performance; however, excluding AI stocks, the S&P 500 actually fell 1% this week, with the number of new lows on the New York Stock Exchange surpassing new highs for nine consecutive days, signaling the near end of "breadth trading." Goldman Sachs bluntly stated that this is a "frustrating cat-and-mouse game" between the stock market and interest rates—any breakout can be snuffed out by the bond market at any time, so equity holders must short US Treasury bonds to hedge simultaneously.
The confrontation between the AI Agent wave and the rise in US Treasury yields is tearing the US stock market into two completely different worlds.
This week, Meta released its Agentic AI model Muse, increasing its market value by $220 billion in just one week and driving the Nasdaq 100 to become the best performing major index. At the same time, US Treasury yields continued to climb, pushing interest rate–sensitive assets into decline—the Russell 2000 significantly lagged, and the S&P 500 repeatedly struggled around the 7700 level, failing to achieve a solid breakout.
The internal divergence in the market has reached an extreme level. If AI-related symbols are excluded, the S&P 500 actually fell by about 1% this week; for nine consecutive trading days, the number of New York Stock Exchange new lows exceeded the number of new highs, marking the longest such streak since October 2023.
Tony Pasquariello, head of Goldman Sachs Global Hedge Fund Business, characterized the current situation as a “frustrating cat-and-mouse game between the stock market and interest rate markets,” and offered a clear recommendation: If you insist on holding stocks long, you must simultaneously short US Treasuries to hedge interest rate risk.

AI Agent Sparks Structural Rally, Market Cap Concentration at Record High
Meta’s Muse model was the week’s biggest market catalyst. The release of this Agentic AI product increased Meta’s market cap by $220 billion in a single week, pushing the total back over $2 trillion, while driving strong performance across the entire AI supply chain—semiconductors led gains due to hardware demand from the “reasoning economy”, as Agentic AI-related stocks benefited collectively. The Agentic AI basket index under Goldman Sachs (ticker GSXUAGNT) rose accordingly, with the high-beta momentum basket recording nine straight days of gains.
Goldman Sachs senior trader Rich Privorotsky pointed out that Muse is only the first of many entrants, with competitors from Google and OpenAI expected to emerge in the coming weeks and months. He believes that as large enterprises take on security, infrastructure, and distribution roles, a lot of economic friction and intermediaries will be eliminated, which itself is a productivity boost and structurally suppresses inflation, having a broadly positive impact on the equity market.
It’s worth noting that concentration in US stocks has reached historic extremes. Goldman Sachs chief technology analyst Pete Callahan noted that there are now ten US TMT companies in the S&P 500 with market capitalizations exceeding $1 trillion, whereas just three years ago, at the start of the AI cycle, that number was only three.

Severe Internal Market Divergence, “Diffusion Trade” Declared Dead
Behind the AI frenzy is a sharp deterioration in market breadth. This week, technology was the only standout sector, while energy and financials lagged; the Mag7 (the seven tech giants) outperformed the other 493 S&P components by almost threefold. ZeroHedge cited data that the “diffusion trade” is effectively dead—NYSE new lows have exceeded new highs for nine consecutive days, marking the longest spell in nearly three years.
Meanwhile, there’s also divergence within the AI rally. AI power-related stocks performed poorly this week, showing a nearly perfect negative correlation with the probability of Democrats winning the House in the US midterm election—there’s concern that if Democrats control the House, current and future data center projects could be put on indefinite hold, posing a potential constraint on US AI infrastructure investment.
In addition, credit spreads for hyperscale cloud companies (Hyperscalers) widened significantly this week, while their equity valuations remained elevated. Goldman Sachs noted that the disconnect between Hyperscaler credit risk and implied Nasdaq volatility has grown to abnormal levels, a gap that merits caution.

US Treasury Yields as Biggest Obstacle, S&P 7700 Faces Repeated Resistance
The continuous rise in US Treasury yields is the key variable holding back a market breakout this week. After trading sideways around 7700 for nearly two months, the S&P 500 showed signs of an upward breakout on Tuesday, but was immediately “hit” by the bond market, falling back to its original level that evening.
Pasquariello described this situation as a “frustrating cat-and-mouse game” between stocks and the rate market: every time the stock market tries to break out, rates jump higher, dampening risk appetite; and as rates rise, tighter financial conditions directly hit rate-sensitive assets. The Russell 2000 has underperformed the S&P 500 for six consecutive weeks, with Goldman Sachs admitting that small-caps continue to be sold by long-term funds and wealth management channels, and the technical outlook has also broken down.
In terms of valuation, research by the Goldman Sachs US equity strategy team shows that the market has already priced in some discount to the sustainability of AI infrastructure profit growth—the median forward P/E ratio of AI infrastructure stocks has dropped to 22x from 32x in April 2026, with some pessimistic expectations absorbed.
Goldman Sachs: Hedge Equity Exposure, Breakout Still Needs a Catalyst
Facing the above fractures, Pasquariello offered a clear operational framework: if you want significant equity long exposure, you should also establish a simple US Treasury short position as a hedge. He rates the current speculative long positioning in the market at +3 on a scale from -10 to +10, suggesting it’s not crowded overall.
He also noted that buyback strength is waning, IPO issuance is rising, and US retail investors are inclined to watch from the sidelines, making the current market structure less than perfect. He also acknowledged that, given the ferocity of recent moves in rates, the market’s ability to hold onto most of its gains is itself impressive.
Pasquariello’s conclusion: if there is an upward breakout catalyst, fast-money funds will have to chase higher quickly; until then, the standoff between stocks and rates will persist, with ZeroHedge adding one last word—“yet.”
Disclaimer: The content of this article solely reflects the author's opinion and does not represent the platform in any capacity. This article is not intended to serve as a reference for making investment decisions.
You may also like
UBS: Maintains Tesla (TSLA.US) “Neutral” rating, target price $385, AI narrative dominates stock price pricing logic
UBS predicts that Tesla's global deliveries in the third quarter of 2026 (3Q26) will be approximately 470,000 vehicles, representing a year-on-year decrease of 5% and a quarter-on-quarter decrease of 1%.
Altcoin Price Targets Point to Six 2027 Levels
The AI frenzy withstands the "5% US Treasury yield"! Nasdaq hits new highs, strengthening the "80/20" pattern as Wall Street reassesses stock-bond allocation
The 30-year US Treasury yield briefly reached about 5.53%, hitting its highest level since 2004, while Brent crude remains above $100 per barrel. Risk assets have once again withstood the pressure, reflecting investors' continued belief that the commercialization of AI applications and corporate profit growth can to some extent offset the impact of rising financing costs and discount rates.
AI infrastructure hits a "wall"! Marvell: Copper interconnects and memory bottlenecks become key to the next stage of computing power expansion
The AI computing power competition is evolving into a system-level game encompassing interconnection and memory. According to Marvell executives, inference models have driven KV cache growth by about 10 times in nine months, and rack-level memory can no longer meet demand. Copper connections are approaching their limits, making optical interconnection and memory expansion core to computing power growth. Customization has become an industry trend, and Marvell is set to benefit from its expertise in analog SerDes, photonic fabric memory, and end-to-end customization.
