AI arms race reaches a crossroads! Goldman Sachs warns that "money spenders" have overvalued, while "money earners" still have room to grow.

```

The debate on the sustainability of AI capital expenditures is playing out in the markets.

Goldman Sachs hedge fund business head Tony Pasquariello warns that this month, core hyperscale cloud computing enterprises have suffered their worst single-month decline since Meta’s IPO. The market is sending a clear signal through prices: the “spenders” have gone too far. Meanwhile, the “earners” of the AI infrastructure supply chain—represented by memory chips—continue to show relative resilience, but cracks in this divergence began appearing at the end of last week.

According to Goldman Sachs' latest research, the core hyperscalers basket has dropped 18% so far in June, marking the group’s worst monthly performance since Meta’s IPO. Pasquariello points out that the market’s message is quite clear: whether large-scale AI capital expenditures can translate into substantial revenue growth will be the key test during the Q2 earnings season at the end of July.

Meanwhile, hedge funds sold US stocks last week at the fastest pace since tariff day last year. Goldman’s prime brokerage data shows that long positions in hyperscalers have fallen to nearly a three-year low, while long positions in AI infrastructure companies are near a three-year high. The internal tension between these positions is fueling continued intraday volatility in the market.

“Spenders” Under Pressure: Hyperscalers Face Worst Month

June has been a painful journey for hyperscalers. Pasquariello bluntly states in the report that the core basket of hyperscalers dropped 18% this month, second only to the impact of Meta's IPO, marking a tough stress test for the core narrative of AI trading.

The market logic isn’t complicated: the entire AI complex is priced on the assumption of ongoing capital expenditure increases and ever-expanding inference demand. However, no one’s forecast models currently incorporate a scenario of “slightly reduced spending.” Pasquariello points out that the breaking point of this system may come the moment a key spender concludes that “spending less” is more beneficial for shareholder returns.

This pressure is also evident at the S&P 500 index level. Hampered by hyperscalers’ heavy weights in the index, the S&P 500 ended Friday down across the board, and has been consolidating for the past month—even as the S&P 500 excluding AI-related stocks and the equal-weighted S&P 500 quietly hit new highs, highlighting clear internal differentiation in the market.

Resilience and Cracks Among “Earners”: Memory Chips Lead, Infrastructure Hit

As hyperscalers came under pressure, the “arms dealers” of the AI infrastructure supply chain once traded in the opposite direction. The US memory chip basket rose 16% in June so far, forming a stark contrast to the hyperscaler declines and reflecting continued market preference toward “earners” in the AI value chain.

Micron Technology’s (MU) strong earnings briefly reinforced this logic—those collecting the AI capital expenditure bills seem to have ample upside. However, Friday’s market action introduced new uncertainty: AI infrastructure names were sharply hit, the Philadelphia Semiconductor Index (SOX) moved noticeably lower, and Goldman’s prime brokerage platform recorded a large influx of supply.

Pasquariello remains cautious about this, but notes a tactical signal worth attention: Goldman Sachs prime brokerage data shows hyperscalers’ long positions are close to a three-year low, which implies that short-term oversold risk may now have accumulated to some extent.

East vs. West: Low-Cost AI Development Shakes Capex Logic

The debate over the sustainability of AI capital expenditures is being reinforced by a new narrative: the divergence between East and West AI development paths.

Pasquariello, referencing Rich Privorotsky’s analysis, points out that if cutting-edge intelligent models can be developed in the East at a fraction of the Western cost, then the West’s biggest capital allocators are also those at highest risk of over-investment. The report mentions that GLM-5.2 was fully trained on 100,000 Ascend 910B processors, without using any Nvidia chips, directly challenging the core assumption of “AI capital expenditures must continue expanding.”

This cost gap between East and West makes questioning the return on hyperscalers’ capital expenditures harder to dispel. Pasquariello admits that just how far this “rubber band” can stretch remains the market’s core unresolved issue.

Positions and Technicals: Short-Term Demand Momentum Weakening

From a technical perspective, Pasquariello sees local conditions becoming more complex recently. Despite last week’s heavy selling, hedge fund net exposure remains at the 87th percentile historically, meaning overall positioning is still heavy.

Month-end and quarter-end asset allocation rebalancing is expected to bring passive selling pressure for US stocks. As June ends, the window for company buybacks will also narrow. Pasquariello notes, the above factors combined mean the demand momentum observed on the Goldman platform will slow, and in the next month the market will rely more on retail capital to absorb shares.

Still, he closes on a relatively optimistic note: historically, the Nasdaq-100 Index has risen in 17 of the past 18 Julys, providing some short-term support for the market.

Risk Warning and DisclaimerThe market has risks; investments require caution. This article does not constitute personal investment advice, nor does it take into account the special investment objectives, financial situation or needs of any particular user. Users should consider whether any opinions, viewpoints or conclusions in this article are appropriate to their own circumstances. Investing based on this article is at your own risk. ```