Goldman Sachs trading desk warns: The rubber band has been stretched to its limit; the market may be approaching a breaking point.

After the plunge in South Korean stock indexes this Tuesday triggered a wave of global sell-offs, Rich Privorotsky, head of Goldman Sachs’ One-Delta trading desk, warned: In recent weeks, the market has virtually ignored all negative signals facing the AI capital expenditure trade, and this rubber band has been stretched to its limit.

On Tuesday, the Nikkei Index fell nearly 3.5%, Korea’s KOSPI index dropped about 10%, and SK Hynix plunged approximately 13% in a single day. Privorotsky noted that trades relating to memory chip stocks seem to be hitting a structural bottleneck. Meanwhile, he emphasized that, as the biggest source of AI expenditure, price movements in hyperscale cloud providers are the core signal most worth tracking.

Technically, the Nasdaq failed to make a valid new high, and support for market structure is weakening. Market makers’ Gamma value is low near current levels and will shrink further as prices fall. CTA strategies are still net buyers by most metrics, but are highly exposed to convexity risk in downward scenarios. The picture presented by Goldman Sachs’ prime brokerage data closely matches price movements: the global market has evolved into an unusually concentrated single bet.

Extremely Concentrated Trade Structure

The sharp decline in SK Hynix is backed by a rapidly expanding leveraged position.

According to reports, the CSOP SK Hynix Daily 2x Leveraged Product (7709.HK) has accumulated assets under management totaling $16.7 billion, making it one of the world’s largest single-stock leveraged ETFs. This very number is the most direct illustration of market concentration risk.

Privorotsky stated in his analysis: "AI is driving the stock market, the stock market is driving economic expectations, and all paths ultimately point to the same few stocks." The market is rationally allocating the gains from AI expenditure to direct beneficiaries—semiconductors, memory, electricity, networking, and infrastructure sectors—this extreme version of the allocation logic has resulted in the rare concentration seen currently.

Overlooked Cost Compression Signals

While pursuing the AI capital expenditure narrative, the market is choosing to ignore another force. Privorotsky pointed out that recent AI technology progress displays clear cost-lowering trends: the rapid iteration of GLM-5.2, the emergence of fused architectures, continued advances in small models, and new progress from Japanese models together form a technical path of sustained cost compression and efficiency improvement.

Privorotsky admitted he was initially skeptical, but when related reports appeared in the MIT Technology Review and were third-party verified by Appen, this signal became hard to ignore.

What deserves particular attention: GLM-5.2 was trained entirely using 100,000 Ascend 910B processors, without any Nvidia chips. This means that if cutting-edge AI capability can be developed in the East at a fraction of the cost required in the West, then the biggest capital spenders currently are also the group most exposed to over-investment risk.

However, the above progress is almost not reflected in the market pricing of hyperscalers’ expenditure expectations. Hyperscalers continue to underperform the broader market, but their willingness for capital expenditure only grows.

When Will the “Breaking Point” Trigger?

Privorotsky raised a sober question about the core logic of this trade: “How far can the rubber band stretch?”

In his view, the triggering condition for the breaking point is always: a major spender concludes—spending a bit less is more favorable for shareholder returns.

The issue is, the scenario of “spending a bit less” is not factored into anyone’s forecasting model. The entire AI industry chain’s pricing logic is based on the assumption that inference demand growth drives capital expenditures to perpetually rise. If this assumption cracks, the valuation system will face reevaluation pressure.

Additionally, the market is systematically overlooking a risk: the highly deflationary pricing power in token economics.

Technical Perspective and Capital Flows

From technical and capital flow perspectives, short-term risks are skewed downward. The positive effects brought by last week’s options expiration have dissipated, and the market now faces month-end and quarter-end rebalancing pressure, theoretically forming directional capital flows of selling stocks and buying bonds.

Problems related to large-scale mega tech companies are becoming increasingly visible, and the Nasdaq failing to make a decisive new high feels like an unstable equilibrium.

In summary, Privorotsky’s judgment is: the current market structure is becoming fragile, concentration risk is severely underestimated, and tracking hyperscalers’ price movements will be the frontline signal to identify whether this trade is breaking down.

Risk Disclosure and DisclaimerThe market is risky, please invest carefully. This article does not constitute personal investment advice and does not take into account individual users’ special investment goals, financial circumstances, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article suit their particular situation. Investing based on this article is at your own risk.