장신 IPO 임박, 한국 국가대표팀 진입: 저장장치 산업 체인의 자금은 어디로 흘러갈까?

Original Author: David, Chaoxiang Research

In the past couple of days, positive news regarding memory has been coming one after another.

On June 29, South Korea launched a semiconductor mega project totaling over 1,000 trillion won (approximately $650 billion), with an official goal to double DRAM production capacity within five years.

At the same time, China's DRAM leader, CXMT (ChangXin Memory Technologies), passed its listing review. The market expects it to be listed between mid-July and early August, with institutions valuing it at 2 trillion to 4 trillion yuan. Combined with the judgment that memory manufacturers will face shortages until 2028, the reasons to be bullish on memory and semiconductors have never been as clear-cut as they are now.

Meanwhile, this sentiment has even spilled over to overseas networks.

Jukan (@jukan05), a well-known tech investment blogger on X, posted that the most worthwhile direction to bet on in the second half of 2026 might still be Chinese semiconductor self-sufficiency targets.

Citing exchanges with Chinese sell-side analysts, he stated that CXMT's market cap after its IPO will be at least 5 trillion yuan. Most of the raised funds will flow into stocks related to domestic semiconductor self-sufficiency, so he believes targets like ACMR (ACMR) and NAURA (NAURA) still have prospects.

However, blindly rushing in on this wave isn't necessarily a good time.

Currently, nearly 30 A-share CXMT concept stocks have a combined market cap exceeding 1.9 trillion yuan. Most leading stocks across the upstream and downstream of the industrial chain are generally trading near their 52-week highs. Blindly charging in is certainly not the optimal solution.

After a round of gains, there are not many segments left in the market where expectations haven't been fully priced in.

Prices Are Up, but Sales Volume Has Barely Budged

In a June 23 report titled "China's CXMT Is Set to Challenge DRAM Incumbents," the US semiconductor research firm SemiAnalysis broke down a set of data:

CXMT's bit shipments in Q1 2026 grew only 11% quarter-over-quarter, but its average selling price (ASP) surged approximately 57% quarter-over-quarter. Bit shipments measure the actual quantity of memory sold based on storage capacity, essentially answering "how much was sold"; ASP measures "how expensive it was sold."

Putting these two numbers together means that during this quarter, CXMT sold barely more units, but sold them at much higher prices.

Therefore, SemiAnalysis concluded that CXMT's recent profit surge relies on the industry cycle itself, rather than technological breakthroughs or market share gains.

In a rising price market, the first to benefit are the original manufacturers directly selling chips—Samsung, SK Hynix, Micron, and CXMT itself. Their profits scale linearly with ASP, and they have been among the biggest gainers over the past year.

SK Hynix's stock price once surged over 350% this year. But at current levels, expectations along this original manufacturer chain have been heavily priced in: Samsung and SK Hynix's current forward P/E ratios are only 3 to 5 times. While seemingly cheap, this reflects that the market has already factored the AI-driven demand and profits for 2026-2027 into the stock prices.

The market has basically acknowledged the profits realized through price increases. The same goes for the batch of A-share memory original manufacturer and module stocks; their gains are already substantial, leaving limited room for further entry. As for whether the upstream expansion chain (such as equipment and materials) has been bought up to the same levels, we'll look at the data later before drawing conclusions.

Overall, I believe CXMT is at a subtle crossroads:

On one hand, it benefits from this cycle by selling DRAM and profiting from price increases. On the other hand, it is the spender, using the 29.5 billion yuan raised from its IPO to expand capacity by purchasing equipment and materials upstream.

In the past month or two, a prevalent sentiment in the market has been to directly "buy CXMT's upstream and downstream." In hindsight, blindly going all-in has indeed yielded significant returns. However, at this current juncture, if one judges that the memory and semiconductor rally will continue, several things must be clarified:

First, where exactly does your target stand in the industry chain, and how does it benefit?

Second, is its current price still at the foot of the mountain, or has it already climbed halfway up or even reached the summit?

Let's answer the first question.

"CXMT concept stocks" is an overused label. Breaking it down, the demand driven by CXMT follows two paths, benefiting different sets of companies and materializing at different times.

The first path is the expansion chain for general DRAM. Currently, 99% of CXMT's shipments are general DDR and LPDDR. Out of the 29.5 billion raised in the IPO, over 22 billion is explicitly designated for equipment procurement related to wafer fabs and technology upgrades. This money primarily flows to front-end equipment (chip-making machines), the largest portion of expansion investment. Once the production lines are running, materials are consumed continuously.

Representative companies in the equipment segment include NAURA (002371), AMEC (688012), Piotech (688072), Hwatsing Technology (688120), and ACM Research Shanghai (688082). Material segment representatives include Anji Technology (688019), Sino High-Tech Materials (300666), Yoke Technology (002409), and National Silicon Industry Group (688126). This chain benefits from the money CXMT is spending right now, offering the highest order certainty.

The second path is the HBM chain, involving a different set of players than the first. HBM is high-bandwidth memory used in AI servers, technically more challenging than general DRAM. CXMT's HBM is still catching up, with production lines expected to start operations only by the end of 2026, lagging behind the general DRAM expansion. Crucially, the value in HBM lies not in front-end etching and deposition but in the packaging stage—stacking, bonding, and molding multiple chip layers. Therefore, the HBM beneficiaries are a different group of companies:

Test equipment provider Jingzhida (688627), packaging material suppliers Huacheng Technology (688535), Lianyungang Zhongfu Lianzhong (688300), Shanghai Xinyang (300236), and advanced packaging and testing companies Shenghe Microelectronics (688820) and Tongfu Microelectronics (002156).

Upstream and Downstream: High Altitude, Thin Air?

Spreading out the targets from the two chains above and ranking them by their current stock price relative to their 52-week high clearly shows where capital has swept through. Data below is as of midday on the 29th.

There's a clear divide in the table.

For general DRAM equipment and materials, almost all are trading near their 52-week highs, mostly within 3% of the year's peak. On June 29, Hwatsing Technology hit the daily limit up, setting a new historical high. Yoke Technology also refreshed its high. AMEC, Anji, and NSIG all saw gains around 10%.

This segment is the market's recognized "shovel seller for CXMT's expansion," with the most solid logic and highest certainty, leading capital to price it most fully. In other words, the certainty of the expansion chain is already baked into the price.

Lagging behind is the HBM packaging segment.

Lianyungang Zhongfu Lianzhong is about 18% away from its 52-week high and closed lower on June 29 against the trend. Shenghe Microelectronics is also about 18% from its high. Packaging and testing company Tongfu Microelectronics is about 9% from its high. The gap between them and the equipment/material segment isn't because they are cheaper or overlooked, but more because their realization timeline is later:

CXMT's HBM production line won't start until the end of 2026. Orders and performance for this group of companies will only truly release after the lines are operational and yields improve. Their lower current positions reflect that "their turn hasn't come yet." Simply viewing them as "bargains" risks incurring time and opportunity costs.

As for the overall conclusion, it's already clear.

"Buying CXMT's upstream and downstream" at this point is no longer a question of getting on board, but whether they are at the summit.

Looking purely at prices, the equipment and materials for general DRAM are mostly at their highest levels in a year. The cheap entry point is gone. The HBM packaging segment is slightly lower, but only if you're willing to wait.

Furthermore, price only tells half the story of "expensiveness." The other half, which price alone cannot articulate, depends on who is buying and selling at these levels.

Hot Money Props Up, Some Are Withdrawing

The essence of this layer is that pricing power has shifted from long-term capital focused on fundamentals to short-term speculative hot money driven by sentiment.

On one side, industrial capital, the National Integrated Circuit Industry Investment Fund (the "Big Fund"), and "national team" funds are systematically reducing positions at high levels. On the other side, retail investors and hot money are rushing in, fueled by the AI theme. The former, perhaps the most knowledgeable about this business, are selling; the latter, aiming to buy low and sell high, are buying.

Let's first look at the withdrawing side, based on a compilation of public information:

• Zhu Yiming, the actual controller of GigaDevice, reduced his holdings by approximately 6.33 million shares between May 11 and 25 (company announcement). He is also the founder and chairman of CXMT. The person most likely to be bullish on the CXMT chain reduced his position in a related leading company at a high point.
• The National Big Fund has been reducing its holdings in NSIG since January, cumulatively cashing out approximately 3.882 billion yuan by early June (company reduction announcement).
• Multiple heavyweight semiconductor stocks, including Montage Technology, Hygon Information Technology, and Piotech, have seen shareholders reduce holdings after their stock prices surged in 2026 (various company announcements).
• The "national team" (Central Huijin) reduced its positions in broad-based ETFs like the CSI 300 (according to Caijing magazine's estimation based on Central Huijin's holdings and fund circulation shares). This reduction isn't specifically against semiconductors but represents a counter-cyclical withdrawal from the overall high market, with semiconductors being the biggest gainer and most obvious candidate for profit-taking in this round.

Source: Fulai Index Investment's Xueqiu column

The motivations for these reductions cannot be generalized. The Big Fund has a divestment cycle; reducing positions upon share maturity to recycle capital is a routine operation. The "national team" adjusting its broad-based holdings could be a counter-cyclical rebalancing rather than a bearish view on a specific sector. Industrial capital and executive reductions have various reasons as well.

Interpreting these moves as a "collective bearish signal" is certainly an overstatement. But one thing is certain:

At the current price levels, these original long-term holders have, almost unanimously, chosen to realize some gains. Regardless of the specific motivation, the action itself conveys that the current price has reached a level where long-term capital is willing to take profits.

Now, look at the buying side.

According to market data cited by Sina Finance, the main force driving this tech rally is hot money:

Northbound capital increased by about 400 billion yuan year-to-date, and margin financing has expanded to roughly 2.8 trillion yuan. This type of money bets on themes and momentum. Whether a stock is near its 52-week high or has a high P/E ratio doesn't affect their entry; they buy the rally itself.

Regulators have also sensed the overheating, applying the brakes by raising margin requirements and suspending trading for consecutive-limit-up stocks for review. Chip ETFs have frequently been suspended due to large premiums (market prices far exceeding net asset value), with premiums once exceeding 30%.

Putting buyers and sellers together, the conclusion is straightforward:

At current levels, long-term capital related to memory and semiconductors is realizing profits in batches, while short-term hot money is taking over. The marginal pricing power here increasingly lies in the hands of capital trading on sentiment.

Note: This does not mean the rally will top out immediately. The memory shortage extends into 2028, and CXMT's expansion involves real capital. However, most institutions believe the sector will digest profit-taking through sharp fluctuations, entering a ranking game that places a higher premium on earnings delivery, rather than a trend decline.

My Personal View

Combining the two layers of judgment with the funding landscape above, I believe the inclination is as follows.

Short-term (around CXMT's listing), sentiment still has momentum. The IPO subscription frenzy and thematic hype on the listing day could give the sector another push higher. However, this is the final leg driven by emotion. The further it goes, the more it becomes a game of passing the parcel, where the risk of chasing highs outweighs the potential reward.

Medium-term (based on CXMT's HBM production line realization), what's truly worth waiting for is the differentiated realization between the general DRAM expansion chain and the HBM chain. Equipment and material orders depend on the speed of capital expenditure deployment after CXMT goes public. The HBM packaging stocks, which are still relatively low, will only have their turn when the production line starts operation and yields improve towards the end of 2026. Getting in now is a bet on timing.

Key Signals to Watch:

1. Turning point in DRAM spot prices. The foundation of this cycle is price increases. Once spot prices turn downward, the original manufacturers and module makers benefiting from price hikes will react first.
2. CXMT's capital expenditure pace post-IPO. Most order delivery depends on these data.