Over the past three months, the U.S. stock market's memory chip sector has surged between 44% and 184%, significantly outperforming the Philadelphia Semiconductor Index's gain of 20% to 88% over the same period, though recent volatility has increased notably. The core market debate centers on whether the sentiment-driven trading enthusiasm can find substantive support from actual earnings during the reporting season.
In a report dated June 24th, JPMorgan Chase maintained its view of a "prolonged and extended upcycle," suggesting recent stock price fluctuations are driven more by sentiment and trading dynamics rather than a deterioration in fundamentals. The market is now entering a concentrated window for verifying actual performance and forward guidance.
Key Verification Points for the Current Cycle
Regarding the quality of earnings in this cycle, JPMorgan Chase focuses on three main verification lines: the pace of Long-Term Agreement (LTA) finalizations, capital expenditure guidance from Cloud Service Providers (CSPs), and structural constraints on the supply side. These variables will collectively determine whether the pricing logic for the memory industry within the AI cycle is entering the next phase of revaluation.
LTA Finalization Pace: A Core Variable for Revaluation
JPMorgan Chase notes that the pace of Long-Term Agreement (LTA) progress is one of the most important structural variables in the current memory cycle. Since May, LTA announcements have noticeably slowed, but this does not signal a trend interruption; rather, it indicates negotiations have entered a more complex stage.
The report indicates that memory manufacturers are becoming more cautious regarding price locks, prepayment ratios, and contract protection clauses, with some corporate governance factors also delaying the signing process. However, the market has seen "sample-level breakthroughs," with the strategic cooperation between Micron and Anthropic being a particularly typical example.
This agreement not only covers long-term supply arrangements for HBM, DRAM, and SSDs but also extends to deep integration at the equity and financing levels. JPMorgan Chase views this as a new type of "LTA-like structure," signifying that supply-demand relationships are evolving from simple supply arrangements towards "capacity + capital" collaboration.
Furthermore, Samsung Electronics and SK Hynix are also seen as key strategic suppliers within the Anthropic ecosystem, further reinforcing the trend of long-term locking in AI memory supply. JPMorgan Chase anticipates that the window for large-scale LTA finalizations will concentrate in the second half of 2026. At that time, U.S. hyperscale cloud providers are expected to become the primary contracting parties, potentially driving the industry into a new cycle of valuation repricing.
CSP Capital Expenditure: Accelerating Share for AI Memory
The second key line stems from changes in the structure of Cloud Service Provider (CSP) capital expenditures. The report shows that the proportion of AI memory within CSP capital expenditures has rapidly increased from less than 20% in 2022 to a projected 52% in 2026, with the potential to further exceed 70% in 2027.
The significance of this shift is that memory is gradually evolving from a "supporting component" of AI infrastructure into one of the core constraint variables affecting the pace of computing power expansion.
However, market divergence around this trend is widening. On one hand, profit expectations for memory manufacturers continue to be revised upward. On the other hand, the market is beginning to question the sustainability of such a high proportion of investment in AI memory, especially against the backdrop of AI commercialization returns not yet being fully realized.
JPMorgan Chase believes this expectation gap will not converge quickly. Historical experience suggests that semiconductor profit revisions often lead the confirmation of capital expenditures, but the ultimate trend still depends on cloud providers' judgments regarding AI server architecture evolution and long-term ROI. Therefore, CSPs' CAPEX guidance during this earnings season will become a key variable determining the intensity of market volatility.
Supply and Technology Paths: HBM Expansion Coexists with "Efficiency Constraints"
The third critical variable comes from structural changes on the supply side. Although major memory manufacturers are accelerating their expansion pace, the industry as a whole still faces significant structural constraints.
On one hand, SK Hynix's DRAM monthly capacity is expected to increase to approximately 630,000 wafers by year-end, and Samsung is also advancing its advanced production line construction. On the other hand, the construction cycle for new, advanced production capacity is as long as 2 to 2.5 years, meaning supply elasticity still significantly lags behind demand expansion.
More importantly, HBM is reshaping the logic of "bit output efficiency." Due to its higher added value and complex packaging structure, an increasing proportion of HBM production can, to some extent, suppress the overall growth rate of bit supply, keeping the industry within a relatively low range of supply growth.
Comments