Since January 2023, driven by AI computing demand, memory chip stocks have experienced a sustained surge, with the top three memory manufacturers' share prices rising by an average of 699%. However, as prices continue to climb, the market's primary concern is shifting from "how much higher can they go" to "what signal should be monitored to identify a turning point." On March 16, the UBS Global Research team published a report titled "Global I/O Storage Semiconductors," reviewing the cyclical patterns of the memory industry over the past 20 years and reassessing current leading indicators. UBS points out that the fundamental logic of the memory industry has changed radically due to the push from AI computing, rendering traditional valuation and forecasting models potentially obsolete. Instead, operating profit has emerged as a more reliable leading indicator. The industry's fundamental logic has changed: AI is pushing supply and demand towards a new equilibrium. UBS attributes the core driver of this market trend to the "upward shift in memory value in the AI computing era." The report highlights two accumulating supply-side constraints behind this revaluation:

High Bandwidth Memory (HBM) is consuming an increasing share of DRAM wafer capacity, leading to a "severe DRAM shortage"; This tightness is further amplified by the "trade ratio": as HBM DRAM die sizes continue to grow relative to DDR, each unit of HBM "consumes" more production capacity.

Building on this, UBS presents a conclusion more sensitive for investors—a rise in the central rate of return. The report states plainly: "We believe ROE has been structurally reset," projecting an average ROE of 36% for Samsung, SK Hynix, and Micron Technology from 2026-2030E, significantly higher than the 15% average over the past decade. This implies that using old cycle templates to identify peaks may lead to more frequent misjudgments. Traditional indicators are failing: The "second derivative" is no longer reliable. Historically, investors often used the "second derivative"—the quarter showing the fastest sequential or year-on-year acceleration in memory contract prices (ASP)—to predict stock price peaks. However, UBS's review shows this indicator's reliability is declining. Out of 10 instances of "stock price peaks" over the past 20 years, stock prices peaked in the same or an adjacent quarter as the sequential change in DRAM ASP only 50% of the time. For example, during Q4 2009 (post-global financial crisis recovery), Q2 2013 (post-industry consolidation cycle), and Q1 2017 (traditional cycle), the peaks in sequential ASP changes occurred 3, 5, and 5 quarters earlier than the stock price peaks, respectively. UBS notes that while the synchronization between actual ASP and stock prices is slightly better (peaking in the same quarter 60% of the time), overall, the "second derivative" has become a difficult signal to rely on for precise "exiting at the peak" in the current complex market environment. Finding a new anchor: Operating profit is a better leading indicator. With traditional indicators failing, what should investors watch? UBS's answer is operating profit (OP). Operating profit reflects not only price changes but also incorporates factors like bit growth and cost reduction per bit. Therefore, it more closely represents the "final value" of the industry's true prosperity. Analysis in the report shows that over the past 20 years, stock prices peaked simultaneously with or before operating profit 90% of the time. Particularly before 2012, stock prices and operating profit peaks were almost synchronous. Since then, the stock market's predictive nature has strengthened, with stock prices typically peaking one or two quarters ahead of operating profit (one quarter earlier in most cases). However, UBS also cautions investors that predicting when operating profit will peak is not easy. The reason remains the structural changes in supply and demand brought by AI, which could make profit timing harder to forecast, especially as HBM continues to squeeze DRAM capacity. The interplay between price, supply, and profit will become more complex, and the estimated timing of the profit peak could shift rapidly. Thus, operating profit can serve as a crucial observation metric, but it is by no means a "panacea." AI is reshaping the industry: Structural ROE reset suggests the uptrend could extend to 2027. UBS emphasizes that the current memory cycle is fundamentally different from previous ones. The advent of the AI computing era has caused a fundamental shift in value towards the memory segment. As HBM occupies an increasing portion of DRAM wafer capacity, the DRAM shortage is becoming more acute. Furthermore, the growing die size of HBM DRAM further intensifies capacity constraints. Based on these factors, UBS believes the Return on Equity (ROE) for the memory industry has undergone a structural reset. The report forecasts that the average ROE for Samsung, SK Hynix, and Micron Technology will reach 36% from 2026 to 2030, far exceeding the 15% seen over the past decade. Consequently, the report maintains an optimistic outlook for memory stocks. It projects that the industry's operating profit will peak in the third quarter of 2027. All else being equal, this suggests the upward trend for memory stocks could persist until the second quarter of 2027. UBS continues to favor SK Hynix as its top buy recommendation, while maintaining "Buy" ratings on Samsung, Micron Technology (MU).