AI is experiencing crazy fluctuations. What are the key indicators to watch for a breakout?

In the past year, AI chip and memory manufacturers have almost consistently outperformed hyperscalers, becoming the strongest theme in global stock markets. However, a report released by JPMorgan on July 1 points out that this continuously widening gap in returns is difficult to sustain in the long term. Whether the AI rally can continue in the future depends not only on capital expenditure but also on whether AI commercialization capabilities can truly be realized.

JPMorgan believes that the AI trade is currently facing two distinctly different evolutionary paths: one in which AI cloud vendors and model companies gradually improve profitability, catching up with the growth of chip companies, thereby driving the expansion of the entire AI industry chain; the other in which excessive profits of chip companies squeeze downstream profitability, ultimately forcing cloud computing vendors to cut capital expenditure, which in turn hits chip demand.

In this process, the market needs to observe not just AI Capex, but indicators such as AI computing power prices and token prices that can directly reflect commercialization capabilities.

Cracks Are Widening in the AI Trade

JPMorgan notes that since last September, the US semiconductor sector—especially AI chip and memory manufacturers—has almost continuously outperformed large cloud platforms, a performance that has reached a level worth caution.

The report argues that a major catalyst for this semiconductor rally is the significant upward revision of capital expenditure expectations by cloud giants.

According to analyst consensus compiled by Bloomberg, the capital expenditure of the five major cloud giants (Google, Amazon, Meta, Microsoft, Oracle) is expected to reach $758.1 billion this year, an increase of approximately 100% year-over-year; by 2027, it is further expected to rise to $925 billion.

For this reason, the AI chip industry chain is seen as the most direct beneficiary of the entire AI investment thesis. But JPMorgan emphasizes that as different segments of the same AI industry chain, such a large divergence in returns between chip companies and cloud giants is inherently unsustainable.

The AI Rally Has Two Completely Different Outcomes

The report believes that this gap will eventually be closed, and there are two possible ways.

The first, and the scenario JPMorgan favors, is that AI commercialization continues to improve.

As cloud giants, AI model providers, and end users gradually enhance the monetization capabilities of AI products, with revenues and profits continuously improving, they will capture an increasing share of value in the entire AI value chain, thereby catching up with the performance of chip companies. In this case, the overall profit of the AI industry chain expands, rather than a redistribution of profits among different segments.

The other is a negative scenario that the market is currently more worried about.

If the excess returns of semiconductor companies are ultimately built on the continuous compression of customer profits, then cloud giants, model companies, and end users may all begin to reduce their willingness to make capital expenditures. Once capital expenditure slows down, it will directly weaken future chip demand, eventually leading to greater adjustment pressure on the previously leading semiconductor sector. JPMorgan points out that this is exactly what many clients are currently worried about.

Why Is the Market Starting to Worry?

The stock prices of cloud giants have been largely flat over the past year. Although profits continue to grow, stock prices have not risen in tandem, meaning that valuations are compressed and the cost of equity financing is actually rising. At the same time, their credit spreads have also begun to be significantly higher than those of semiconductor companies, indicating that the cost of debt financing has also increased.

Another data point that worries the market comes from capital expenditure forecasts. According to market consensus, the growth rate of cloud giants' capital expenditure will slow down significantly starting in 2027. It is expected to grow 100% year-over-year in 2026, but only 22% in 2027, and then drop to single digits in subsequent years.

Although JPMorgan's own forecast is more optimistic than the market's, expecting Capex to reach $1.15 trillion in 2027, higher than the market consensus of $925 billion, if the current market forecast materializes, the entire AI trade could face sustained adjustment, affecting not only the stock market but also the credit market.

What Really Deserves Attention: AI Computing Power Prices

In JPMorgan's view, the most important indicators to watch in the future have already changed.

The report points out that the key to determining the future profitability of cloud giants is no longer just the scale of capital expenditure, but the price of AI computing power. The higher the price of computing power leasing, the easier it is for cloud platforms to increase AI service revenue, and to maintain or even expand their profit margins, thereby supporting continued large-scale capital investment.

The report shows that AI computing power prices had been under downward pressure until the end of 2025, but there was some improvement from April to May this year, only to decline again in June. At the same time, the token prices of large language models (LLMs) rose steadily from the end of February to the end of May, then also cooled down in June.

JPMorgan believes that these two price indicators can directly reflect the progress of AI commercialization and are core observation variables for judging whether the AI investment thesis can be sustained in the future.

Risk Warning and Disclaimer

        Markets are risky, and investment requires caution. This article does not constitute personal investment advice, nor does it consider the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment based on this is at your own risk.