The structural trend within the AI technology sector remains intact, yet the question of "where to sell" is increasingly preoccupying a significant number of investors. As the ChiNext Index, represented by core themes like the "Xin-Ning Combination" (referring to the high-growth narratives of "AI-driven price increases" and "overseas expansion in new energy"), continues its ascent, holders face a practical dilemma: should they sell incrementally on the way up, or wait for a definitive signal before exiting? In a recent research report dated May 21st, a strategy analyst from Guotou Securities, Lin Rongxiong, offered a relatively clear perspective. Based on the "N-shaped" pricing pattern typical in technology investments, the so-called "selling point" (Point D) is more accurately understood as a liquidation point rather than a point for partial profit-taking. The strategy of "selling on the rise" often backfires in practice—investors frequently buy back shares during the uptrend, inadvertently increasing their exposure instead of reducing it. Underpinning this framework is a more fundamental logic:
The peaks of leading stocks in an industrial trend almost invariably form an M-shaped double-top pattern. The first peak is driven by trading sentiment and is notoriously difficult to predict in advance. The second peak, determined by fundamentals, represents the true actionable selling opportunity.
The report establishes a stark quantitative benchmark: investors should become vigilant when a company's stock price has already priced in its earnings growth for the next three years. If the price discounts growth for the next five years, that represents the physical limit for an A-share tech stock's rally. Compared to the brutal collapse of the dot-com bubble in 2000, today's AI giants are far from that precipice, primarily because they hold substantial cash reserves. The true danger zone may not materialize until around 2027. The M-Top Rule: The First Peak is Unavoidable, the Second is the Real Opportunity Reviewing past industrial trend cycles reveals a remarkably consistent pattern where leading stocks ultimately form M-shaped double tops.
The 2020-2021 New Energy Wave: Contemporary Amperex Technology Co. Limited (CATL) hit its first peak (trading peak) in December 2021, a time when its fundamentals were sound. The true fundamental peak occurred between June and August 2022, coinciding with the peak in quarterly TTM profit growth.
Kweichow Moutai formed its trading peak in January-February 2021 (while liquor prices were still rising), with its fundamental peak arriving in July-August 2021 (as high-frequency liquor prices began to turn), a gap of approximately five months.
The 2013-2015 Tech Bull Market: The ChiNext Index reached its trading peak in June 2015, with the fundamental peak following around December 2015 to January 2016, an interval of about half a year.
The 2000 Nasdaq Dot-com Bubble: The trading peak was in March, with the fundamental peak (marked by downward revisions in downstream capital expenditure and Cisco's earnings outlook) in August-September, a 5-6 month gap.
The pattern is clear: the two peaks are typically separated by 1-2 quarters. While the first peak is nearly impossible to foresee, the second peak—corresponding to the peak in TTM profit growth—is a signal that can be tracked. Taking CATL as an example, the ideal cautious point based on the PE overvaluation framework would have been August-October 2021 (a left-side signal), while the true fundamental selling point was June-August 2022. The report notes that investors who successfully exited that cycle shared a common trait: their selling points were biased to the left, essentially reflecting a willingness in their investment thesis to forgo the final stretch of gains. Three Years of Discounted Growth is the Warning Line, Five Years is the Endpoint Understanding this valuation framework starts with simple arithmetic.
A company with a current PE of 200x would be considered reasonably valued if it can achieve an annualized profit growth of 85-100% over the next three years, which would translate to a forward PE of 25-30x. However, if the three-year growth rate is only 30%, the implied forward PE remains around 90x, a clear danger signal.
Calculations show that among A-share companies since 2007 (excluding recent IPOs and those with market caps below 5 billion RMB), only 40% have achieved three-year annualized growth above 30%. A mere 17.44% achieved growth above 50% over three years, 5.44% above 30% over five years, and a scant 1.15% above 50% over five years.
Sustained high growth is inherently a low-probability event. If a stock price already demands the company deliver growth rates of 50% over three years or 30% over five years to justify its valuation, the margin for error approaches zero. Historical A-share industrial trend leaders confirm this ceiling: CATL's 2021 peak discounted 3-4 years of growth; Kweichow Moutai, 4-5 years; BYD, 3-4 years; WuXi AppTec, 3-4 years; Mindray Medical, approximately 4 years; Inovance Technology, over 4 years.
Most leaders peaked at valuations discounting 3-5 years of growth. The 2000 dot-com bubble serves as a counterexample—companies like Microsoft and Cisco were priced for growth far exceeding five years at their peaks, leading to a subsequent "double whammy" of valuation compression and earnings downgrades. In contrast, during the 2000 bubble, overseas leaders like Microsoft, Cisco, Apple, and Amazon were priced for growth far beyond three years. Subsequent significant earnings downgrades led to a collapse in both valuation and earnings, bursting the bubble. This lesson serves as a crucial reference point for assessing current AI tech bubble risks. Based on this, a simple guideline for long-term valuation centers emerges: the steady-state valuation for A-share companies over the medium to long term should generally not exceed 30x PE. Current AI Leaders Are Not Yet in the Danger Zone, but Close Monitoring is Needed Around 2027 The report emphasizes that, according to Wind and Bloomberg consensus estimates, current stock prices for both domestic and international AI leaders discount growth for less than three years, not yet breaching the warning line. A three-layer assessment paints the following overall picture:
The Nasdaq 100's PE of 36.5x places it at the 91.4th historical percentile—high, but far below the 206x seen in 2000, positioning it at the border between "safe" and "watch"; The Capex/Operating Cash Flow ratio is around 70% and rising rapidly, placing it in the "watch" zone; The average cash coverage ratio is 94.4%, but Meta's has fallen to 37.3%. It is projected that by 2026, free cash flow for several companies besides Microsoft will turn negative, raising uncertainty about the sustainability of capital expenditure into 2027.
The most fundamental difference from 2000 lies in the cash flow structure. The dot-com bubble burst when infrastructure players (like Cisco) eroded the cash flow of downstream internet companies—Amazon and others relied on financing to burn cash, providing orders for Cisco. Once interest rates rose or the economy weakened, this funding cycle broke. Currently, tech giants like Microsoft and Google still possess robust operating cash flows. Their massive capital expenditures are supported by genuine cash generation, not merely piled on through financing. However, the report also cautions that "robust" describes the current state, not a permanent guarantee. If capital expenditure continues to expand in 2026-2027 while AI commercialization fails to meet expectations, a deterioration in cash flow structure will likely be the earliest warning signal. Crowding: 37% May Not Be the Peak, but the High-to-Low Rotation Index is a Practical Signal From the perspective of institutional positioning crowding, the peaks of the last three industrial waves varied: the "Mao Index" (high-quality blue-chips) peaked at 45% positioning around 2019-2021; the broad new energy theme peaked near 40% around 2020-2022; AI tech positioning currently stands at approximately 37%. Given the unique characteristics of each industrial cycle, it is ineffective to simplistically use a specific positioning level as a peak signal—conditions are never identical. In dealing with positioning crowding, market practice often relies more on the "A-share High-to-Low Rotation Index." When this index reaches elevated levels, the market often exhibits a strong endogenous momentum to rebalance towards lower-positioned sectors. Current data indicates that the divergence between high and low-positioned sectors has not yet reached an extreme. There is no immediate urgency for a systematic style rotation, and the tech sector's relative advantage persists. However, as the divergence gradually widens, attention should be paid to whether extreme levels are triggered subsequently. As the market cap of AI leader InnoLight Technology surpasses the trillion-yuan mark, the演绎 chain of the four-stage theory for tech investment is advancing: from giants (ChatGPT, NVIDIA, Microsoft) → infrastructure investment (computing power) → key产业链 links (AI chips) → application端 with supply-demand gaps. Supply-demand gaps fall into two categories:
Upstream: Memory, fiber optics, electricity/power, energy storage, and copper等, where the core logic is price increases. Downstream: Robotics, autonomous driving, AI ecosystem software等, where the core logic is application penetration.
Drawing an analogy to the 2021 "Ning Combination"行情, the lithium mining sector saw far greater gains and institutional inflows than the battery makers themselves. By 2026, the pricing focus for AI tech should gradually shift towards sectors with supply-demand gaps, where price increases are inevitable. This view has already been validated by the recent strong performance in the memory sector. Subsequent directions such as robotics, copper, and power/power grid merit continued attention.
Comments