BlockBeats News, July 20: US Stock AI Trading is Transitioning from "Buy the Rumor" to "Show Me the Return".

Over the past week, global semiconductor stocks faced a sharp sell-off. The Philadelphia Semiconductor Index fell into a technical bear market range from its recent high, with some AI chip, storage, and equipment stocks retracting by 20% to 30%. Previously crowded and favorable trades suddenly became the market's pressure release valve.

There were many surface reasons triggering the adjustment: TSMC's capital expenditure increase raised concerns about the return on investment in AI, investors reevaluated chip stock valuations, soaring storage prices brought "chip inflation" pressure, geopolitical risks and oil price rebound heightened macro uncertainty.

But there is only one deeper issue: AI capital expenditures have become so large that they can no longer rely solely on narratives; the market needs to see evidence of monetization.

JPMorgan and Morgan Stanley, in two reports released on July 20, approached this market trend from two angles.

JPMorgan focused on hyperscalers, such as Google, Amazon, Microsoft, and Meta. The bank believes that the AI cycle is still intact, but the core of trading is shifting from upstream hardware to midstream platforms. In the upcoming earnings season, the market will focus not only on revenue and EPS but also on AI capex guidance, data center rollout pace, and whether management can explain how these expenditures translate into cloud revenue, advertising efficiency, enterprise AI services, and free cash flow.

According to JPMorgan's calculations, AI-related capital expenditures may reach nearly $870 billion by the end of 2026, with a year-on-year increase of about 77%, of which hyperscalers will contribute approximately $750 billion. Such a scale of investment has exceeded the imagination of the traditional tech investment cycle. Investors who were willing to pay a premium for "compute scarcity" in the past are now starting to demand proof from companies: can the money spent ultimately form a profit pool?

This is also the key to the recent market style change. Previously, the massive gains in upstream AI hardware led to highly crowded momentum trading. Once the market begins to worry about the capital expenditure slope, funding pressure, supply expansion, and end-game profit margins, semiconductors are naturally the first to be sold off. JPMorgan's assessment is that the mid-term risk-return may be better positioned in AI midstream, namely in cloud providers and some software security assets. If there are more specific AI monetization clues on future earnings conference calls in the coming weeks, funds have the opportunity to rotate from AI hardware back to mega-cap quality growth.

However, Morgan Stanley's attitude towards memory stocks is significantly more optimistic. The bank made it clear in the report's title—The Sell-off in US Memory Stocks Creates an Attractive Entry Point.

Their reasoning is that this memory cycle is unique. Past DRAM and NAND cycles were mainly driven by PC, mobile, and traditional server inventory, but now data centers and AI represent the primary demand. Morgan Stanley's channel checks indicate that data center memory prices rose by at least 25% from the second quarter to the third quarter, exceeding both their own and third-party expectations. DRAM saw a nearly 70% sequential increase in the first quarter and over 40% in the second quarter. Given these figures, a slowdown in the rate of increase is considered normal, and interpreting it solely as a sell signal can easily lead to misjudging the cycle.

The market is concerned about long-term contract price suppression, NVIDIA reducing some memory specifications, and storage manufacturers increasing capex leading to new supply. Morgan Stanley does not ignore these risks. However, the bank emphasizes that these factors are more likely to alter the cycle's shape, smoothing out the price and profit peaks, but potentially extending the cycle's duration. For stocks, a gradual climb in earnings over a few years often supports valuation more easily than a windfall profit in a single year.

Moreover, Morgan Stanley believes that memory has become one of the core bottlenecks in the AI industry chain. AI spending could grow at over 50%, the complexity of HBM4 will absorb more capacity, new platforms like Rubin Ultra may continue to increase memory content, and there is synchronous growth in demand for low-power DDR5, enterprise storage, and more. In other words, while supply will increase, the rate of AI's consumption of memory is also rising.

Looking at both reports together, Wall Street is not simply bearish on AI at the moment. What has truly changed is that the market is starting to reassess the AI industry chain.

In the first phase, investors were buying into scarce computing power, with GPUs, HBM, advanced packaging, EDA, and network equipment all receiving valuation expansions. In the second phase, investors are beginning to ask who can control bottlenecks, who can pass on costs, and who can convert capital expenditure into revenue and cash flow. JPMorgan leans towards hyperscalers in this logic; meanwhile, Morgan Stanley believes the memory shortage is not over, offering an improved risk-reward ratio post-correction.

This makes the upcoming earnings season particularly crucial. Alphabet will be the first to report on July 22, followed by Microsoft and Meta on July 29, and Amazon on July 31. Nvidia, Broadcom, and Micron will then release their results from late August to September. The market will be watching three key points: whether capex continues to rise, if AI revenue can be quantified, and whether bottlenecks such as memory, GPUs, and power will continue to drive up unit costs.

If cloud providers offer stronger AI monetization evidence, the recent semiconductor sell-off may be seen as a healthy rotation, with AI trades transitioning from hardware elasticity to platform profitability. If management continues to emphasize investment scale but fails to provide a path to returns, the market's patience with high-valuation tech stocks will continue to wane.

The most delicate aspect now is that the chip stock decline has not erased the AI narrative but has made this storyline more tangible. In the past, the market only needed to believe in a shortage of computing power; now it also needs to factor in the price, duration, financing cost of the shortage, and ultimately the payer.