According to Wall Street, citing a comment by Rich Privorotsky, the head of the Goldman Sachs One-Delta trading counter, the recent global stock market lags behind and the hot trade around AI capital spending has shown signs of vulnerability. The core judgement is that the market has neglected almost all negative signals over the past few weeks, and that funding is increasingly concentrated in a few AI beneficiary shares.

The transaction is concentrated on a few AI objects

After the South Korean stock market collapsed on Tuesday, the pressures were rapidly transmitted to the global market. The mantra dropped close to 3.5 per cent that day, the Korea KOSPI index fell about 10 per cent, and the SK Hercules dropped about 13 per cent a day. In Goldman Sachs ' view, this is not only a single-day fluctuations, but also a reflection that transactions related to the storage of chips are close to structural bottlenecks.

Privorotsky notes that the current market has a highly single pricing path. AI gains from expenditure were concentrated on the semiconductor, storage, electricity, network and infrastructure sectors, which eventually further shrunk to a few lead stocks. Goldman Sachs' large brokering data also show that the global market has evolved into a highly concentrated bet.

  • The menstrual index dropped by 3.5%.
  • Korea KOSPI index drops by about 10%
  • SK Hercules dropped about 13% a day.

Leverage product magnification volatility risk

The article mentions that behind the SK Hercules fall is the element of rapid accumulation of leveraged funds. According to the report, the management of the SK Hercules 2-trigger product, 7709.HK per day, has increased to $16.7 billion and is among the largest single equity leverage ETF in the world.

According to Goldman Sachs, the rapid expansion of these products is in itself a sign of increased concentration risks. Once the core has fallen, leverage may increase price volatility and increase market re-pricing of similar AI transactions.

Cost reduction signals are not fully valued

Privorotsky also mentioned that the market, in pursuing AI capital expenditure narratives, ignored that technological advances were lowering the cost of model development. The iterative GLM-5.2, the progress of small models and the emergence of new structures all point to a cheaper and more efficient direction.

In particular, he noted that the GLM-5.2 allegedly completed training on the basis of a $100,000 turbo 910B processor and did not use the British Weeda chip. If front-line AI capabilities can be achieved at a lower cost, then the super-large cloud service providers that are currently most active in capital may face the challenge of overinvestment.

Key variable is cloud service provider expenditure

Goldman Sachs believes that what really needs to be targeted is not a single chip unit, but rather the performance of a super-large cloud service provider. Since such companies are the main source of AI capital expenditure, the valuation base of the entire chain may be affected once either of them judges that “less money” is more favourable to shareholder returns.

In terms of the trading structure, the NASDAQ index has not reached a decisive new high, and the support after the expiry of the option has weakened. With pressure to rebalance at the end of the month and the end of the season, the market may face changes in the flow of funds to sell stocks and buy bonds. According to Goldman Sachs, the current AI transaction is in a precarious balance, and if cloud service providers expect to spend less, the market may be more revalued.