An analysis of 10 historical demand-driven super cycles reveals that nine of them formed double tops, with the average drawdown from the first peak to the trough being 22.1% over 5.1 months, followed by an average rebound of 44.8% from the trough to the second peak.
The success of a second peak depends critically on whether commodity prices can remain strong, whether inventory levels can tighten again, and whether earnings expectations can continue to be revised upward. In 2018, MLCCs experienced a sharp reversal due to inventory accumulation, order cancellations, and capacity expansion. In contrast, 2021 memory chips formed a double top supported by inventory depletion and supply discipline.
Current AI memory leaders have experienced an average drawdown of around 40%. Historically, the forward P/E near the first trough generally does not significantly exceed the valuation midpoint at the cycle's start. The domestic forward P/E has already returned closer to the cycle's starting point earlier than overseas counterparts, suggesting improved odds. However, on the probability side, the adjustment has only lasted 23 trading days, indicating that positions are still in a concentrated exchange phase. The supply-demand, inventory, and earnings dynamics have not yet formed a re-acceleration loop. The sharp sell-off at the end of July, followed by a rebound the next day, more closely resembles a "shock-style clearing." Confirmation of the first trough and a second leg still requires waiting for right-side signals such as a stabilization of fundamentals, a halt in valuation decline, and a recovery in price and volume. The current environment offers improved odds but lacks high probability.
Why the Baseline Scenario Points to a Double Top, and the Band Returns from First Trough to Second Peak Are High
A review of 10 historical cycles 鈥?including fiber optics, uranium, iron ore, rare earths, mobile phone memory, MLCCs, lumber, container shipping, photovoltaic silicon, and lithium 鈥?shows that nine formed a pattern of "first peak, a phase of significant decline, then a second peak." Only the 2018 MLCC cycle failed to produce a clear second peak. The average maximum drawdown after the first peak was 22.1%, lasting an average of 5.1 months. The average gain from the first trough to the second peak was 44.8%, ranging from 27.4% to 77.8%. As of July 29, 2026, the average drawdown for current AI memory leaders is about 40%, a depth that has entered a historically comparable range, but the adjustment has lasted less than a month, meaning the time dimension has not yet fully cleared.
The Key Differentiator Between a Double Top and a Sharp Reversal Lies in Commodity Price Persistence and the Inventory Cycle Turnaround
The first stock price peak typically leads the peak in commodity prices, while the second peak is closer to the commodity price inflection point. A necessary condition for the second peak to form is that commodity prices remain strong during the decline from the first peak. In the early cycle, the monthly return correlation between stocks and commodities is about +0.32, but by the middle-to-late stage before the first peak, this correlation approaches zero. From the first peak to the first trough, average stock returns turn negative while cumulative commodity returns remain positive. From the first trough to the second peak, both generate positive returns again. Only after the second peak do they both turn negative. The 2018 MLCC cycle formed a sharp peak due to rising channel inventories, order reductions, and capacity expansion. The 2021 memory cycle formed a double top after inventory depletion, supported by real demand, manufacturer capital expenditure discipline, and a re-acceleration in prices and earnings. Whether the current AI cycle can form a second peak depends on whether HBM/DRAM supply-demand dynamics more closely resemble the 2021 memory cycle or the 2018 MLCC cycle.
Odds: Forward P/E Near the First Trough Historically Does Not Significantly Breach the Cycle Start Valuation, and Chinese Tech Cycle Stocks' Valuations Have Returned to the Cycle Start Earlier Than Overseas Peers
Super cycles typically transition from a first leg driven by expectations to a second leg driven by earnings. The typical path is "valuation decline, continued upward earnings revision." Only after valuations have been digested and earnings continue to materialize can stock prices switch from being expectation-driven to earnings-driven. Currently, the AI upstream sector has moved from a phase of "rising valuations and earnings" into a valuation digestion phase ("first peak to first trough"). According to Visible Alpha consensus earnings estimates, as of the first half of 2026, the median forward P/E for the Chinese sample rose by approximately 48%, and earnings expectations were revised up by about 11%. For the overseas sample, the forward P/E rose by about 23%, and earnings expectations were revised up by about 25%. From June 30 to July 27, the median forward P/E for the Chinese sample fell by about 42%, while earnings expectations remained largely flat. The overseas P/E fell by about 16%, and earnings expectations were still revised up by about 1%. Currently, the forward P/E of Chinese tech cycle stocks has fallen back to the cycle's starting point earlier than their overseas counterparts. The median forward P/E (Visible Alpha consensus) for the domestic sample is now 1.18 times its cycle low point, while overseas is at 1.54 times. Confirmation of favorable odds for a second peak comes from the P/E stabilizing near the cycle start, while confirmation of high probability comes from continued upward revisions in earnings expectations.
Probability: Earnings Persistence, High Dividend Yields, and New Narratives Are Worth Tracking, But None Are Present Yet
After the first peak, the market becomes noticeably insensitive to current earnings reports. There is no stable positive relationship between a single quarter's profit beat and the stock's post-earnings performance. Only multiple consecutive quarters of earnings beats can increase the credibility of a second earnings-driven leg. High dividend yields are typically a lagging result of peak profits, high payout ratios, and low stock prices. In five historical cases, the peak dividend yield occurred on average about 15 months after the first peak. Dividends can cushion a decline but are not necessarily sufficient to drive a trend-following valuation recovery.
The Overall Direction of the Industry Chain Generally Follows the Core Product, but Equipment and Downstream Segments Show Significant Peak-to-Trough Dislocations
The lithium battery industry chain showed the strongest synchronicity, with lithium carbonate as the core product. The monthly return correlation coefficients between representative companies for lithium iron phosphate, electrolyte, lithium hexafluorophosphate, and lithium battery equipment and the representative company for the core product were 0.56, 0.55, 0.77, and 0.75, respectively. However, copper foil and PVDF failed to form a second peak due to supply release and individual company expansion pressures.
Current Memory is Closer to a 'Shock-Style Clearing Candidate'; the First Trough Cannot Be Confirmed Yet
Public fund holdings concentration data shows that by the second quarter of 2026, the dispersion of AI positions had exceeded that of the 2021 lithium battery cycle. However, while positions in early-stage directions were declining, positions in memory, PCBs, and optical modules were still increasing. A synchronized reduction in positions across the entire industry chain has not yet occurred. Historically, the first trough has primarily followed two price-volume paths: "a contraction-driven stabilization" or "a panic-driven volume spike followed by a reversal." Observing the current memory situation, it appears closer to a "panic-driven volume spike sell-off, followed by a volume-driven reversal the next day" 鈥?a candidate for shock-style clearing. On July 29, a core memory company fell 9.9%, only to rebound 18.4% on July 30, closing back above its 5-day moving average. The 7-day and 5-day average trading volumes were 1.13 times and 1.52 times the average of the previous six months, indicating a concentrated exchange of panic sellers and buyers. However, a pattern of higher lows and a repair of the 10-day and 20-day moving averages have not yet been completed.
Overall Assessment: Supports the Double Top Pattern, with Odds Entering a Historically Attractive Zone, but Probability Awaits Right-Side Confirmation
Historical precedent, current valuation levels, and the depth of the drawdown collectively suggest that the initial phase of valuation expansion and trading congestion has been largely released. The marginal benefit of continuing to be bearish based solely on valuation compression is diminishing. The conditions that are not yet met primarily revolve around three dimensions: time, positions, and fundamental re-acceleration. The memory adjustment has only spanned 23 trading days so far, significantly shorter than the average clearing period of about 5 months seen in historical samples. Trading volume and turnover remain high, indicating that positions are in a phase of concentrated exchange rather than natural exhaustion. Furthermore, HBM/DRAM prices, customer inventory levels, supply yields, and earnings expectations have not yet formed a virtuous loop of "inventory re-tightening, price maintenance, and earnings re-acceleration." Subsequent judgments should proceed in the order of "the fundamental gate, the valuation anchor, and price-volume confirmation." On the fundamental side, focus on monitoring new HBM/DRAM supply, customer inventory levels, cloud manufacturer capital expenditure, and the spread between long-term contract and spot prices. On the valuation side, observe whether the forward P/E can stabilize at low levels and whether earnings expectations stop being revised down. On the trading side, wait for core companies to first not make new lows for 2-3 consecutive trading days, then reclaim their 10-day and 20-day moving averages, form a pattern of higher lows, and finally show declining volume during pullbacks.
Key Risks
Key risks include global AI capital expenditure falling short of expectations; overly rapid supply expansion in AI upstream sectors like memory; intensified friction between China and the US in technology, trade, and finance; weaker-than-expected domestic policy support, implementation effects, or economic recovery; a sharper-than-expected tightening of domestic and overseas macro liquidity; and further escalation of regional conflicts such as those in Russia-Ukraine and the Middle East.
Comments