The author of this article is | Freddy Data support | Gugu Big Data (www.gugudata.com) 01

In January 1995, the film "A Chinese Odyssey" premiered. In the film, Stephen Chow's character, Joker, delivers a classic line to Athena Chu's character, Zixia, that has influenced two generations: "I once let the truest love slip away from before my eyes, only to regret it after losing it. The most painful thing in this world is nothing more than this. If God could give me another chance, I would say three words to that girl: I love you. If I had to put a time limit on this love, I hope it would be... ten thousand years!" Stephen Chow and Athena Chu, a talented man and a beautiful woman, missed each other on screen, and in reality, after twists and turns, they ultimately missed each other as well. It wasn't for lack of love. It was because there was too much love. In the vast, bustling world, temptations are everywhere. With Stephen Chow's youthful and impulsive self-control at that time, how could he have been determined to love only this one person, even though Athena Chu was considered stunningly beautiful at the time?

The capital markets are exactly the same. Just like AI, its innate beauty and standing head and shoulders above the crowd are as obvious as lice on a bald head. However, it cannot withstand various human weaknesses such as lack of understanding, fear of heights, and hesitation. Countless investors have watched this AI giant ship gradually sail away. Even given N more chances, they would still miss it.

02

Because the gains in the first half of the year were too substantial, and coinciding with the repositioning of fund styles at the end of June, over the past seven trading sessions, almost all AI-related sectors have experienced declines of approximately 20%. For a sector representing the ultimate industrial revolution and the future of humanity, a 20% drop is quite significant! Could this be what Joker meant: if God could give me another chance?

Then, today, the adjustment ended, and a more violent upward surge began! By today's market close, about 2,400 stocks rose across the entire market, while approximately 2,900 fell. The STAR 50 Index led the gains among major indices. The Shenwan Semiconductor Index surged over 6% in a single day. Among the top 30 ETFs by gain in the entire market, the terms "chip" and "semiconductor" dominated the list. ETFs focused on the three sub-sectors of semiconductor equipment, memory chips, and advanced packaging all exploded, almost all hitting their daily limit-up.

Each ETF represents an index, and each index is composed of a batch of constituent stocks. These ETFs hitting limit-up means the indices they represent, and the constituent stocks within those indices, basically all hit limit-up! Such market action has never occurred in the history of the A-share market! This historic-level market movement is not accidental. Following the resonance between the super-cycle in memory and the major cycle of domestic computing power, domestic semiconductors, acting as the "water sellers" of the industry chain, are entering the best golden growth phase in history.

Driven by both the "volume" of capacity expansion and the "price" of technological iteration, coupled with the long-term logic of continuously increasing localization rates, the dual rise in performance and valuation for China's AI industry has only just begun. In the first half of the year, all carbon-based industries—whether real estate, banks, brokerages, or even the relatively appealing innovative drugs—were largely suppressed and struggled, while the silicon-based AI industry stood out alone.

Many people are unconvinced, saying it won't stay like this forever. Just like countless bubbles in history, this time will be no different. It's just sector rotation. A single sector advancing alone is bound to create a bubble and is doomed to fail. The problem is, AI is not just a sector; at least, it's not a sector for carbon-based people. It fundamentally doesn't need carbon-based industries as teammates, so naturally, it cannot be described as advancing alone. It can form a closed loop by itself, and it is highly profitable. It is generating wealth at a terrifying speed unprecedented in human history! Look at Anthropic!

Of course, for those stubborn investors, no amount of talking will help. They still won't buy AI; they will continue to buy baijiu, food and beverage, tourism, real estate... Over seven trading sessions, the entire sector corrected by 20%. The opportunity was given to you; you didn't cherish it. That's your problem.

03

The collective and violent surge in the domestic A-share semiconductor sector today is due not only to the sector's immense underlying energy and sufficient adjustment magnitude but also to another very important reason: ChangXin Memory Technologies (CXMT) is preparing for an IPO.

As the domestic leader and the only IDM achieving scaled mass production of DRAM, market expectations for its post-IPO valuation start at 5 trillion RMB. More optimistically, expectations have been raised to the $1 trillion USD level (approximately 7 trillion RMB). With a bit more optimism, it heads towards $1.2 trillion USD, roughly corresponding to a forward P/E ratio of around 20x.

This might seem a bit expensive because companies like SK Hynix, Samsung, and Micron trade at roughly 6-8x. However, since AI has become a national policy in China, and memory is no longer viewed as a cyclical stock but a growth stock, given CXMT's irreplaceable infrastructure position within China's AI ecosystem, a 20x multiple is not really excessive.

This is not merely thematic speculation but a value re-rating supported by fundamental industry shifts. It will comprehensively activate the growth potential of domestic computing power from multiple dimensions including supply, cost, and ecosystem. The trillion-dollar valuation expectation has solid industrial and performance support—the following details about CXMT's basic operations can be skipped if you're not scrutinizing every detail:

"In fact, CXMT's profitability inflection point arrived more forcefully than market expectations: In 2023, the company incurred a loss of 19.2 billion RMB. In 2024, the loss narrowed to 9 billion RMB. In 2025, it successfully turned profitable, achieving a net profit exceeding 7 billion RMB. Entering 2026, with the memory super-cycle overlapping with capacity release, H1 revenue is estimated to reach 110-120 billion RMB, with net profit attributable to shareholders around 60 billion RMB, a year-on-year increase exceeding 2000%. Its capacity utilization rate has increased from 85% in 2022 to 95.7% in 2025, basically operating at full capacity.

More crucially, CXMT's value lies not only in general-purpose memory but also in its positioning within the core track of AI memory. Currently, the company has completed sample delivery for HBM3 and initiated mass production preparation for 12-layer stacked HBM, planning to shift part of its total DRAM capacity to HBM production. At a time when AI computing power's demand for high-bandwidth memory is exploding, entering the high-end memory track directly opens up the long-term growth ceiling, shifting its valuation logic from a cyclical product to an AI growth product.

For the domestic computing power industry, the significance of CXMT aiming for a trillion-dollar market cap extends far beyond the birth of a leading company. It fundamentally breaks the core bottleneck that has constrained the development of domestic computing power for many years, creating comprehensive industrial stimulus. First is the sense of security from supply chain autonomy, thoroughly clearing supply obstacles for computing power expansion. In the past, the large-scale deployment of domestic computing power clusters always faced the 'chokepoint' problem of high-end DRAM supply: the overseas big three controlled capacity allocation, with shipment schedules and product specifications subject to their control, potentially becoming an invisible shackle restricting the expansion of domestic computing power. In the DRAM field, CXMT's current monthly capacity is over 200,000 wafers, far below Samsung's over 600,000 and SK Hynix's over 500,000. Domestic manufacturers' global market share is only about 5%, leaving vast room for substitution. After CXMT's large-scale mass production, domestic computing power manufacturers will obtain stable, controllable memory supply, fundamentally eliminating supply disruption risks. The expansion speed of domestic computing power is expected to accelerate comprehensively.

Second is the direct benefit on the cost side, leveraging investment enthusiasm in computing power infrastructure. Currently, CXMT's memory chip costs are 15%-20% lower than those of overseas original manufacturers, and incremental capacity is prioritized for domestic manufacturers. As AI servers' requirements for memory capacity and bandwidth continue to increase, the proportion of memory in server BOM costs has risen from 20% to over 30%, becoming a core cost item in computing power construction. The proliferation of domestic chips will directly reduce the initial investment in AIDC construction and computing power leasing, optimizing project profitability models, thereby attracting more social capital into computing power infrastructure and accelerating the penetration of AI computing power from large internet companies to various industries.

The deeper stimulus lies in the resonance across the entire industry chain,打通 (打通) the ecological closed loop of domestic computing power. CXMT's capacity expansion is not an isolated event. Upstream, it will accelerate the domestic verification of semiconductor equipment and materials. Domestic equipment manufacturers like Topsec, AMEC, and Naura have deeply integrated into its supply chain, with order scales continuously expanding. Downstream, it will promote the maturation of memory module and domestic server whole-machine solutions. Module manufacturers like Jiangbolong and Biwin Storage obtain stable supply sources, further enhancing the cost-performance ratio of domestic servers.

Finally, there's the upgrade in technological synergy, pushing domestic computing power towards high-end攻坚 (攻坚). HBM is the core配套 (配套) for high-end AI computing power. CXMT's breakthrough in the HBM field will create a synergistic effect with domestic AI chips, solving the chokepoint pain point of high-end memory power. This will propel the domestic computing power industry from catching up to leading."

Setting CXMT's target at a trillion-dollar market cap signifies the first time a Chinese capital market company would reach a trillion-dollar valuation (PetroChina's brief surge in 2007 doesn't count). Its essence is the market's concentrated pricing of the value of domestic computing power autonomy. It marks that China's semiconductor industry officially possesses the capability to compete on the same stage as global giants. It also ensures CXMT can obtain sufficient funds for capital expenditures, further driving the entire upstream and downstream sectors. Orders will be so abundant that semiconductor equipment manufacturers won't be able to handle them all.

More importantly, positioning CXMT at a trillion-dollar market cap opens up valuation space for all companies in the domestic AI industry chain! Reflected in the capital markets, it means that within the AI industry chain, if you don't buy, it will most likely keep rising day after day to show you.

04

There's another short-term catalyst: SK Hynix's ADR listing on the U.S. stock market tomorrow. Previously, because funds wanted to buy ADRs in the U.S., a large amount of capital directly sold Hynix's underlying shares in South Korea. Over the past ten trading sessions, the stock once fell by 32%. Once the ADR lists tomorrow, this short-term suppressing factor will disappear. Its likely decent performance will bring stronger catalysts to the global AI sector and the Chinese AI sector.

This Friday, global memory leader SK Hynix's ADR is set to list on the U.S. market. For global tech capital, this is not just a cross-market listing for one company; it means the global AI industry chain will welcome a new pricing anchor. As the global leader in HBM (High Bandwidth Memory) technology, Hynix is a core beneficiary of this wave of AI memory. The company firmly occupies the core track of AI memory, with a forward P/E for 2026 expected to be less than 7x and less than 4x for 2027, showing exceptionally significant valuation advantages. With the formal U.S. ADR listing, the logic for value repair continues to strengthen, and the stock price still has upward momentum. In the future, it will create a new valuation anchor point for the global memory sector.

Precisely because the market has been reluctant to give Hynix a growth stock valuation, it hasn't formed a typical bubble. Put more professionally: Hynix currently resembles a skepticism-driven bull market driven by EPS, not a valuation bubble driven by P/E expansion. In fact, the real judgment framework isn't whether a 6x P/E is cheap or not, but whether the high profits from HBM over the next two to three years can transition from cyclical profits to structural profits. I believe, of course, they can. In that case, a 6x P/E is severely undervalued. In fact, if Hynix is disproven, memory is disproven. If memory is disproven, AI is disproven. Is that possible?

For A-shares, the listing of Hynix's ADR not only verifies the strength and sustainability of the memory super-cycle but will also directly lift the valuation center of the domestic memory industry chain and semiconductor equipment manufacturers, pushing the market trend to deeper levels. On one hand, domestic semiconductor equipment and materials companies continue to make breakthroughs in the field of advanced packaging. Segments like CMP, Hybrid Bonding, wet cleaning, and TGV (Through Glass Via) are becoming important beneficiaries of the next round of industrial upgrading. On the other hand, sub-sectors like silicon photonics, optical communication, and high-speed connectors are also beginning to benefit from the accelerated construction of global AI data centers.

At the current juncture, for A-shares, the listing of Hynix's ADR will most likely add more fuel to the already blazing domestic computing power market trend. If the market's focus in the past was "who can produce the strongest GPU," then in the coming years, the market may be more concerned with "who can solve the connection, packaging, and bandwidth bottlenecks between GPUs." From this perspective, the opportunities in domestic semiconductors are also shifting from single-point breakthroughs to industry chain resonance. With the continuous expansion of AI capital expenditure, areas like advanced packaging, semiconductor equipment, key materials, and optical communication are expected to become important main themes in the next round of market trends.

05

Epilogue: One question: Will domestic AI stand at the center stage (C位) of the new global AI cycle?

The recent structural divergence in A-shares continues to play out, with the market repeatedly confirming the core conclusion through round after round of price action: carbon-based sectors are mostly experiencing阶段性 (stage-based) rebound repairs, while AI is the absolute main theme贯穿 (running through) the industrial cycle. From the market cap of optical modules surpassing that of baijiu leaders to the全线爆发 (all-out爆发) of semiconductor equipment and memory chips, the trend of capital concentration towards AI hard tech is becoming increasingly clear.

The core anchor point of this market trend is precisely the comprehensive breakthrough of the domestic AI industry chain: CXMT's IPO targeting a trillion-dollar valuation,补齐 (complementing) the final短板 (short board) of the computing power foundation. Hynix's ADR listing adds more fuel to the fire.

In the past, within the global AI wave, global capital never placed mainland China's AI industry at the center stage (C位). Instead, they looked at Taiwan, China's chip foundries, South Korea's memory, and the U.S.'s large model applications. But in reality, none of the above countries/regions possess China's top-down national destiny and policy坚定性 (firmness), nor the completeness of the AI industry chain and the diversity of application落地 (landing) scenarios. Given time, domestic AI will most likely stand at the absolute center stage (C位) of future global AI trends, moving from following to running alongside and leading. The爆发 (爆发) of the domestic AI industry has actually only just begun.

As Dr. Gelong, founder of Gelonghui, stated at the mid-term strategy meeting: Ordinary people, don't miss this once-in-a-millennium opportunity. Away from the era, you are nothing! As long as you顺应 (comply with) the major trends of the times, the major潮流 (tides), and锚定 (anchor onto) the era's标的物 (target assets), you will earn money you never dared to imagine in this lifetime. If you still don't understand what I'm saying, one, go back and review Stephen Chow's line again; two, study Gelonghui's chart below, which divides the eras since the founding of the PRC into 4.5 periods, showing the涨幅 (gains) of the锚定标的物 (anchored target assets) in each era.

▍ Past Reviews Gelonghui Statement: The views expressed herein are solely those of the original author and do not represent the views or立场 (position) of Gelonghui. Special reminder: Investment decisions should be based on independent thinking. The content of this article is for reference only and does not constitute any actual operational advice. Trading risks are borne by the individual.