百年一遇的内存短缺：苹果涨价、韩国扩产，资金流向哪里？

Original author: Jia Liu, Yuesheng Beatz

This round of memory shortage has finally reached consumers.

On June 17, Cook was still complaining to the *Wall Street Journal* about cost pressures. He said that when consumers need devices, supply has decreased, and memory manufacturers are passing huge price hike pressures downstream. Memory pricing and supply must return to levels that consumer products can bear.

Less than a week later, on June 25, Cook again spoke to the *Wall Street Journal*, describing this cost shock as a "once-in-a-century flood." He said that in his over 40-year career, he had never seen anything like it in any field. Musk, who had just become the first person in human history with a net worth exceeding one trillion dollars, quickly echoed on social media, calling it one of the most violent price jumps he had ever seen.

Then he announced Apple's price increase.

Looking back, these two interviews seemed more like a public relations铺垫 for Apple's price adjustment.

Apple's explanation was straightforward: it wasn't that they wanted to raise prices, but the rapid expansion of AI data centers had driven up demand for memory and storage chips. Previously, the company had tried to avoid passing costs onto consumers, but could no longer bear the burden.

The timing was delicate. The day before Apple announced the price increase, Micron had just released an astonishing earnings report: gross margin of 84.9%, revenue up fourfold year-over-year, and its stock price surged in after-hours trading. The following day, Apple's stock price plunged, wiping out over a hundred billion dollars in market value. Upstream memory manufacturers reaped cyclical windfalls, while downstream consumer electronics manufacturers began passing the bill to consumers.

This is the true transmission chain of this round of AI infrastructure expansion: cloud giants scramble for GPUs and HBM, data centers compete for electricity and servers, memory manufacturers lock in long-term contracts and raise prices, and eventually Macs, iPads, Xboxes, and potentially future iPhones pay the price.

And Apple is not alone. Microsoft also announced it would raise the price of Xbox consoles starting August 1, with the 512GB model increasing by $100 and the 1TB model by $150. The reason given is that console storage and memory prices have already risen by more than 2.5 times and could double again by fall 2027.

But Micron doesn't quite agree with Apple's reasoning.

Micron's Chief Business Officer, Sumit Sadana, didn't directly name Apple, but he told the *Wall Street Journal* that some customers in the past were very aggressive on pricing, and the entire industry shut down a lot of investment plans in 2023 due to low prices and low margins. Tom's Guide directly interpreted this as Micron insinuating that major customers like Apple have been suppressing prices for a long time.

Thus began a war of words between Apple and Micron: Apple says "I don't want to raise prices, memory is too expensive"; Micron argues, "You squeezed prices so no one wanted to expand capacity; now there's a shortage, don't blame the suppliers entirely."

As users, consumption growth is inevitable, so we need to make up for it elsewhere: what will be the next investment trend following this memory shift?

Micron Locks in Profits, Korea's Twin Giants Forced to Up the Ante

To understand the severity of this shortage, first look at the earnings report just released by Micron, a competitor to SK Hynix and Samsung.

For Q3 of fiscal 2026, Micron reported revenue of $414.56 billion, compared to just $93.01 billion in the same period last year; non-GAAP EPS of $25.11; gross margin of 84.9%. The Q4 guidance is even more staggering: median revenue of $500 billion, gross margin around 86%, median EPS of $31.

This is no longer the gross margin structure of a typical cyclical company.

More important than the earnings report are the long-term contracts Micron holds. The company has signed 16 Strategic Customer Agreements, typically covering 2026 to 2030.

Among these, 14 agreements are based on minimum prices, with cumulative revenue of approximately $100 billion. Micron also expects to receive around $22 billion in customer deposits or financial commitments. The agreements are full of take-or-pay clauses, minimum purchase quantities, and price floors, with some even including price ceilings. Micron emphasizes that even with price range agreements, the floor prices can generate gross margins higher than any cyclical peak in history.

Regardless of the quarrel between Apple and Micron, the most significant signal from this war of words is directed at the two Korean giants.

If they don't expand capacity, the most profitable orders in the coming years will be locked in by competitors; but once they massively expand capacity, they will have to pay for the next cyclical downturn. As a result, both Samsung and SK Hynix have entered an investment race.

At yesterday's press conference, South Korean President Lee Jae-myung announced plans to invest 800 trillion won to build four chip factories, with Samsung and Hynix each constructing two new facilities. South Korea is betting its next national industrial position on the hardware gateway of AI.

Semiconductors are already a nation-defining industry for South Korea. Exports, conglomerate profits, exchange rates, employment, and stock market valuations are all tied to Samsung and SK Hynix. During the previous electronics cycle, South Korea benefited from global digitalization through memory and smartphones. In this AI cycle, South Korea isn't competing for application gateways; it's vying for the HBM, advanced packaging, and memory supply that every AI server from Nvidia, AMD, Google, and Microsoft relies on.

General investors can think of HBM as a high-speed memory stack next to AI chips. AI training and inference require not just GPUs but also continuous data supply from memory. Whoever can get certified earlier and deliver faster will secure the most scarce orders in this AI infrastructure cycle.

This is why Samsung and SK Hynix are at the table.

The most critical variable here is the time lag. Expansion isn't about investing money today and shipping goods tomorrow. SK Group Chairman Chey Tae-won has already stated that the memory shortage could last until 2030 because adding new wafer capacity takes at least four to five years. HBM is even more challenging than regular DRAM; it involves not just front-end wafers but also TSV, thinning, stacking, bonding, advanced packaging, and testing.

In other words, the money South Korea invests today addresses supply for 2027, 2028, and 2029. It may not immediately relieve the current price pressures faced by Apple, PC manufacturers, and server manufacturers.

After Expansion, Which Sectors Will Capital Target?

With expansion confirmed, we need to look at the beneficiaries of the expansion narrative. Orders for building these memory factories first flow to equipment, materials, packaging, facility management, and power systems. They will receive the money first.

Yesterday, Yoke J KJ hit the daily limit, and the market has already given a signal.

Yoke J KJ is a supplier of upstream materials to major memory manufacturers like SK Hynix. The market is not betting on it making HBM, but on its position in the material segment preceding HBM and DRAM expansion. As long as SK Hynix, Samsung, and Micron increase wafer starts, consumables like precursors, specialty gases, CMP slurries, photoresists, and wet chemicals will follow suit.

This is why a memory cycle usually doesn't stop at Micron, Hynix, and Samsung themselves, but quickly spills over to the "memory upstream."

Expansion primarily corresponds to front-end fab orders. EUV and DUV lithography, deposition, etching, cleaning, ion implantation, CMP, metrology, and inspection—every step needs to be rescheduled.

In the US and European markets, ASML corresponds to lithography and EUV; AMAT (Applied Materials) covers deposition, materials engineering, and CMP; LRCX (Lam Research) corresponds to etching, deposition, and cleaning; KLAC (KLA) corresponds to inspection and metrology; TER (Teradyne) and COHU (Cohu) correspond to testing; MKSI (MKS Instruments) and ICHR (Ichor Systems) correspond to process subsystems like vacuum, power, and gas/liquid delivery. These stocks performed much stronger than memory stocks last night in the US market.

If mapped to the A-share market, capital typically seeks out the domestic semiconductor equipment chain: NAURA Technology (北方HC), AMEC (中微GS), Piotech (拓荆KJ), ACM Research (盛美SH), Hwatsing (华海QK), Kingsemi (芯源W), Jingce Electronics (精测DZ), and others.

The logic isn't that they all directly supply Samsung or Hynix, but that global memory expansion boosts equipment sector sentiment, while expectations for domestic expansion at ChangXin Memory Technologies (CXMT) and YMTC (Yangtze Memory Technologies) and import substitution are re-evaluated simultaneously.

Materials and consumables are also areas most easily repeatedly explored by capital in this cycle.

Equipment orders are one-time, but material consumption is continuous. As long as fab lines are running, every wafer consumes precursors, electronic specialty gases, photoresists, wet chemicals, CMP slurries and pads, sputtering targets, silicon wafers, quartz parts, and filtration materials. HBM is more complex than standard DRAM, with more process steps and higher material consumption intensity.

The core reason Yoke J KJ was hyped is precisely this. The market interprets it as an upstream reflection of SK Hynix's expansion: Hynix's HBM volume ramp-up drives demand for memory wafer starts and high-end process materials, giving precursor and other material suppliers room for re-pricing.

Under the same logic, A-share capital will also look for targets along several lines: Precursors and electronic materials (Yoke J KJ, Nata GD (南大GD)); Electronic specialty gases (Wateer QT (华特QT), Jinhong QT (金宏QT)); CMP materials (Anji KJ (安集KJ), Dinglong GF (鼎龙GF)), etc.

In the overseas materials chain, ENTG (Entegris) corresponds to filtration, chemicals, materials handling, and CMP-related consumables; LIN (Linde) and APD (Air Products) correspond to electronic specialty gases and industrial gases; GLW (Corning) can serve as an indirect indicator for glass and materials supply chains, but it's not a pure memory materials target.

The elasticity of such companies hinges on three questions: whether they have entered the supply chains of major manufacturers, whether the material value per wafer can increase, and whether price increases and volume growth can truly be reflected in gross margins.

Other beneficiaries include import substitution and Chinese memory capacity.

If the conflict between Apple and Micron continues to escalate, terminal manufacturers will more actively seek alternative supply. While CXMT and YMTC are not perfect substitutes for Samsung, Hynix, and Micron, they can become new bargaining chips in supply chain negotiations.

As long as domestic DRAM, NAND, and HBM capacity is re-priced, A-share equipment and material suppliers will be sought after by capital again. GigaDevice being added to the DRAM ETF is the best illustration.