SDIC Securities: This wave of AI technology is more likely to be M top rather than A kill
The Zhitong Finance App learned that SDIC Securities released a research report saying that up to now, in late June, it was verified that the highs and lows formed by phased overheating have been verified. There are still reservations about a clear top, and the current wave of AI technology is more likely to be M top rather than A kill. At the same time, judging from the three-kill framework of “killing emotion, killing valuation, and killing logic,” the murder mentality and murder valuation have been clarified, but the killing logic is still unconfirmed. Finally, what I still want to emphasize is that the so-called “general will not get off the horse”, that is, before the industry trend is over (the degree of implementation of the AI industry chain's earnings report for the second quarter), and before the two obstacles, the macro gray rhinoceros (strict domestic regulation and overseas entering a cycle of aggressive interest rate hikes) and the collapse of the industry competition pattern (overcapacity in computing power), the market will continue to return to the main line of industry trends. Keep in mind that this point is unbreakable.
The main views of SDIC Securities are as follows:
Over the past two months, a series of research and forward-looking research on “technology selling points” has led to this major discussion on “technology selling points” in the entire market. I am very thankful that this “throwing bricks and leading to jewels”. In particular, the in-depth observation of M Top is deeply rooted in the hearts of the people, and based on this, it is proposed: don't get tangled up, do two, and stick to the top. According to the recent disclosure of the Fund's second quarterly report, it can be seen that TMT's holdings are over 60%, and Pan-AI technology holdings in this round have significantly surpassed the historical Maoning Group level. Combined with the recent sharp decline in AI technology, quite a few people in the market think that they have clearly broken down at the transaction level. The “AI Technology Group” is disintegrating, and the M top 1, which is based on the final end of the industrial wave, is clear.
Originally, the SDIC strategy launched the “Technology Selling Points” series of studies in late May, and only wanted to explain two points: one was that AI technology gradually became overheated in stages based on the A-share high cut and low index in late June; second, it was more inclined to gradually overheat AI technology in stages on the left side; second, it was more likely to prefer high and low stages on the left side, rather than the top of M being clear. With the start of Meta rental computing power dispute in late June, to Apple's Micron dispute, to the US-Iran conflict once again driving up expectations of interest rate hikes, to K3 triggering the theory that the rapid expansion of capital expenditure of leading companies directly caused free cash flow to negative for the first time since listing. Combined with the sharp fluctuations in the Korean stock market in July and the sharp retreat of the US Philadelphia Semiconductor Index, these huge macroeconomic and industrial disputes continued. In July, AI Technology's stock price dropped sharply, with the biggest drop of more than 20% in the core index represented by Double Innovation. Here, I would like to further restate a few core ideas:
1. If you don't have technology, you won't be good. It can be seen that since the 924 market, the market has risen from 2,800 points to 4,000 points, and AI technology has contributed at least 700 points. At the same time, the fact that capital markets protect high-quality development is overwhelming. Based on itself, 2026 has always been compared to 2021 rather than 2015. Other than dividends, it is more difficult for low-level sectors to systematically undertake technology spillover funds.
2. A peak is difficult to predict. It's more about looking back, but it vaguely and effectively corresponds to the highest profit position in a single quarter, and the deviation will generally not exceed one quarter. Obviously, the market's attitude towards “burning money for growth” has changed from unconditional support over the past three years to careful questioning of returns. From the perspective of research and perception, the current huge controversy is repeated at the macro level or is biased in terms of industry. It is more defined as naturally sensitive to negative information after the stock price reaches a high level, and it is still not clear that it is related to the ultimate damage to industrial logic. Judging from Mao Ning, the top of M is vague and effectively corresponds to the highest profit growth rate of leading companies in a single quarter, while the market mostly expects AI leaders to have the highest profit growth rate in a single quarter or in the first half of 2027.
3. Every time A falls sharply, it will not be exactly the same as the previous one. Some varieties will never come back after falling, and the varieties that have continued to rebound may not be the ones that have fallen the most. This means that even if the next main line is AI, the internal core track is likely to switch again. There is a considerable probability that AI core pricing will gradually shift from a “price increase” signal to a “volume increase” signal after this high cut.
4. According to the “big - small - big - small (strong alpha)” rule: the three stages of trend - group - crazy, corresponding trend speculation (the market first recognizes industry trends, capital flows into the most deterministic leader) - Group speculation (as industry logic is widely accepted, subsequent capital spreads from the leader to small and medium-sized companies throughout the industry chain) - finally insanity corresponds to the top of the core leader of a large company (the most important attribute of the top is the top where the leader exceeds the second or third tier) - the top of M is the top of the top three (2) Strong line alpha). Take the big, small, big, and small stages of the new energy industry chain in 2020-2022 as an example. The 2020 new energy trend began in Ningde → spread to the lithium battery industry chain and auto parts before October 2021 → the crazy stage after October, the Ningde era climbed to the top with second- and third-tier strong alpha varieties in 2022.
Since reviewing the AI market since 2023, there have been seven rounds of retracement. It can be seen that this round of AI market retracement was not the biggest. The AI index fell by 47.3% and the optical module index fell by 40% during the period. After that, catalytic driving indices such as Sora's release, Kimi's first exit, the 9/24 financial policy combo, and DeepSeeRK-1 recovered their previous high in February 2025. In this process, AI completed investment from hot topics to industrial prosperity. transformation. This round is the 7th round of retracement, corresponding to a 20.2% decline in the entire AI industry chain and a 32.6% decline in AI hardware from June 30 to July 28, 2026. It took a total of 28 days (20 trading days), and the previous high was not recovered until July 28. What needs to be clarified is that the core characteristic of this round of retracement is that AI hardware is falling fast and fast. According to the three-kill framework of “killing emotion, killing valuation, and killing logic,” it now shows clear characteristics of killing sentiment and killing valuation based on extreme crowding and congestion. The valuation system shows a trend of returning to the mean, and in the context of K-type differentiation without a clear convergence, it has always been thought that it is still not “killing logic.”
Further observed in conjunction with the M peak, the “kill mood, kill valuation, and kill logic” essentially corresponds to the three-level progression of the industry trend market from chip issues to price issues to value issues, and each occupies a clear position in the M top structure — killing sentiment (drop 15%-20%) occurs in the middle of the industry trend before the top, and is a phased high and low level on the left; killing valuation (falling 30%-40%) is more likely to correspond to a bubble return around one peak (trading top); killing logic (falling 40% or more) occurs in the middle of the industry trend before the top (fundamental) What followed was the destruction of value. More specifically:
1. As for homicidal sentiment, it is essentially a chip problem: the company hasn't changed, the industry hasn't changed, and the valuation is not high; it's just that trading is crowded after the market rises to a high level, and profits are settled in stages due to excessive overheating. “The market is a bit panicked.” The driving factors include a sharp decline in peripheral markets, sudden negative rumors, pre-holiday capital avoidance, short-term tightening of liquidity, and extreme sentiment transmission, etc., but they are all unrelated to corporate profits. The pattern is mainly a sharp decline. The magnitude is sharp but the duration is short, usually bottoming out from a few days to a few weeks. The empirical measure is that high and low retracement on the left side generally does not fall below the 120-day EMA, and the core index generally does not fall below the 120-day EMA. The corresponding profit forecast on the profit side is not changing or even improving. The repair path is rapid V-shaped recovery, and the market returns to the main line of the industry.
Historically, the sharp decline in the Ning portfolio in March 2021 (the biggest drop of 8.1% in the Shanghai Composite Index and close to -21.6% in the GEM index) is a typical example: after the decline, a bear market did not open, but rather the Ning Group replaced the Mao Index's clear main line switch. Essentially, it was a structural adjustment rather than a systematic decline. Since then, the Ning portfolio has risen 17.5% from February 18 to November 30. In terms of actual combat response, the correct action during the killing phase was “not to move” — not to blindly lower the position, hold on to the main line, etc., and return to the main line, etc., all of them were educated by a V-shaped backlash in March 2021.
2. When it comes to killing valuation, it is essentially a price issue: the company hasn't changed, the industry hasn't changed, but the price is too expensive — the driving factors include monetary policy tightening, market style changes, performance growth falling short of expectations, and the collapse of sector capital groups. There is a high probability that it will form an M top, corresponding to the return of the mean value after the valuation bubble. The quantitative scale is “overdraft is reduced for three years, and the overdraft is completely clear for five years”. The technical signal fell below the 120-day EMA and turned downward. The core index fell by more than 30%. The valuation behavior moved closer to the valuation center, and the stock price overdraft of performance returned to within 2 years. At this point, the fundamentals did not deteriorate. If it were at the top, the profit side was vague but effectively corresponds to the highest position of profit growth in a single quarter (the degree of deviation will not exceed one quarter).
Historically, Maotai's profit grew as usual in 2021, and Ningde's 2021 Q4 growth rate reached a record high, so what killed the valuation drop was not performance; it was the illusion that “it will take a high growth rate in the next 5 years to be fulfilled” in the price. The temporal characteristics of upward valuation adjustments usually last several months or even half a year. In terms of structural characteristics, internal differentiation occurs. Only overvalued sectors/stocks are killed. Undervalued varieties are relatively resistant to falling, and time is needed in exchange for space — either to absorb overvaluation through performance growth, or stock prices fall to a position where valuation matches performance. After stabilization, it will usher in structural repair, but it is difficult to quickly return to the previous high point. If it is clear that the top of M is one top, then there is an interval of 1-2 quarters (most of the historical samples were 5-8 months), one top retracted at least 20%, normal 30%, and the top rebounded at least 20% after bottoming out. Ideally, the rebound of the top two on the road to repair can return to a position 10% to 15% lower. The actual response is not to get tangled up with one top and do two. The focus of the process of facing the second top after retracting from one top is to adjust the position structure rather than increase the position, and switch the position from the leader of the core strong beta to the second- and third-tier strong alpha.
3. As for the logic of killing, it is essentially a value issue: when the second one arrives, the company changes, and the value is gone. The trigger factors were the introduction of disruptive industrial policies, the replacement of technological iterations, the permanent contraction of downstream demand, the collapse of the competitive pattern of industry, the reversal of supply and demand, and the rewriting of business models, etc., leading to the double death of Davis in response to the systematic downturn in profit forecasting and the restructuring of the valuation system. The morphological decline was extremely deep (more than 40%), and the duration was extremely long (in terms of years, for example, when new energy fell for three years), the rebound was very weak, and “fell and fell”. Technically, it showed a rebound, but the “two tops were 10%-15% lower than the top” and then broke again. The valuation behavior was that the valuation center moved systematically downward, and growth PE converged towards mature PE. After falling, some varieties never came back. The profit side judging measure is that TTM's high profit growth rate has passed, and the consistent expected batch downgrade (over 60% of the downgrades is a strong signal) — the current high increase in earnings reports is only a rearview mirror; the inflection point of high-frequency data and the consistent downturn in expectations are the road signs. The actual response was “clearing the inventory and not going back”: the two players actually left the field and refused to break the bottom.
As can be seen, the deep decline in the A-share structural sector usually follows the progressive path of killing mood → killing valuation → killing logic: in the beginning, panic triggered a sharp emotional decline, then the mid-term valuation bubble bursts into a return to valuation, and finally, if fundamentals are falsified, it will evolve into the worst kind of logical kill. Conversely, rising markets also often correspond to the process of emotional repair → valuation increase → logical verification. From the perspective of retracement and decline, it can be seen that killing sentiment and killing valuations often fall sharply, while killing logic often declines slowly or fluctuates after two peaks. This has been verified quite consistently in terms of domestic and US stock pricing:
1. At the domestic level: Kweichow Moutai retraced from a peak of 2,601 yuan on February 10, 2021, to a low of 1970.01 yuan (-24.26% from the top) on March 10, hurriedly and violently; then, on June 7, the second peak of 2,271 yuan (87% of the top) formed a standard M peak. Since then, the 2021 H2-2022 growth rate has decelerated and the valuation continued to decline, and the killing logic gradually unfolded. The top of the Ningde era was 688 yuan on December 2, 2021, to a low of 368.5 yuan on May 9, 2022 (-46.44%, clearly exceeding the standard by 30%, amplified by systemic shocks such as the Federal Reserve's aggressive interest rate hike and Russian-Ukrainian conflict); on June 24, the second peak was 563.5 yuan (82% of the first top), which was weak; the rebound was weak and the trend was downward after 2023. The so-called three tops and four tops were more of a pulse-type rebound. The same rhythm also occurred in Ning Group: Ning Group fell by around 20% in a single month at the beginning of 2022; instead, the late 2022 kill logic showed a fluctuating decline; after the stock disaster in June 2015, the GEM index plummeted, then rebounded by more than 40%, then slowly declined after the fundamentals peaked at the end of 2015.
2. Overseas level: The 2000 Nasdaq Tech Network bubble provided the most complete overseas sample for M Top and Three Kill. March 2000 was a trading selling point. It vaguely corresponds to the highest profit growth rate in a single quarter in the 1999Q4 NASDAQ index. After the NASDAQ index plummeted, it rebounded again to a high level in August/September 2000 was a fundamental selling point, and the profit growth rate was the highest in the 2000Q1 TTM scale. On the eve of its peak, there are four major left-side warning signs: 1. Industry signals, from “growth myths” to “profit warnings” — Synopsys drastically lowered revenue expectations on February 18, 2000, and the star stock MicroStrategy fell behind due to accounting issues; 2. Macro signals, liquidity pumping out of salary — the Federal Reserve raised interest rates 6 times in a row from June 1999 to May 2000, and interest rates rose from 4.75% to 6.5%; 3. Market sentiment, irrational fanaticism peaked — the average return on the first day of the 1999 IPO reached 7,3% Internet IPOs reached 89%; 4. In a landmark event, confidence completely collapsed — on April 3, 2000, Microsoft was convicted of violating antitrust laws. The stock price plummeted 15%, and NASDAQ plummeted 7.6% in a single day, becoming the last straw to crush the market. It is worth noting that the “Cisco order slowdown” rhetoric that the market relishes occurred at the end of 2000, so the real “killing logic” and weakening fundamentals actually occurred in 2000Q4, not during the first round of sharp decline.
So far, in late June, it was verified that the highs and lows formed by phased overheating have been verified. There are still reservations about the one top. As for the current wave of AI technology, it is more likely to be M top rather than A kill. At the same time, judging from the three-kill framework of “killing emotion, killing valuation, and killing logic,” the murder mentality and murder valuation have been clarified, but the killing logic is still unconfirmed. Finally, what I still want to emphasize is that the so-called “general will not get off the horse”, that is, before the industry trend is over (the degree of implementation of the AI industry chain's earnings report for the second quarter), and before the two obstacles, the macro gray rhinoceros (strict domestic regulation and overseas entering a cycle of aggressive interest rate hikes) and the collapse of the industry competition pattern (overcapacity in computing power), the market will continue to return to the main line of industry trends. Keep in mind that this point is unbreakable.
Risk warning: The opinions in the article do not represent investment suggestions, data measurement errors, and historical experience does not necessarily represent the future.