Memory module prices plummeting, memory stock prices falling across the board, has the memory super cycle peaked?

Wall Street Headlines

In the near term, spot prices for DRAM have plunged sharply, with shares such as those of Micron also taking a major hit. Analyst Dan Nystedt believes this is a signal that the traditional cycle is peaking, because smartphone makers are unwilling to accept the high DDR4 prices. However, analyst Jukan pushed back, saying that the price resistance is limited to traditional memory; DDR5 and HBM demand remain strong, and memory companies are no longer “cycle stocks.” HSBC also believes the market’s current concerns are overstated. The AI-driven memory supercycle is only at the midpoint, and the shortage could last for one to two years.

In the recent period, after memory prices had been rising for months, they suddenly reversed lower, sparking worries in the market that the memory cycle may be topping out.

According to market tracking data, across the U.S., multiple retailers have seen broad price cuts on DDR5 memory. The maximum drop for a single set reaches as much as $100. Taking Corsair’s VENGEANCE line as an example: the 32GB model with a top frequency of 6400MHz is currently priced at about $379.99, down sharply from a recent peak of $490—an over-$110 drop for a single set.

The domestic market has also been hit. A wholesaler told China Business News that the price of mainstream 16G memory sticks “dropped more than 100 yuan in a single day,” and large buyers who had stockpiled earlier are疯狂ly dumping inventory.

“Starting last Saturday, prices just collapsed.” Wang, a wholesaler who has run a storage equipment business in Bai Nao Hui for years, told the outlet candidly. He showed an extreme price chart for a mainstream 16G 3200MHz memory stick: in May last year it was only just over 130 yuan. Then it went on a狂 surge to a peak of 980 yuan in December. But after months of volatility at high levels, the current spot price has already fallen to around 700 yuan.

Wang said helplessly that because the price increases have overdrawn consumer expectations, “they won’t buy unless it’s a necessity. Compared with before November last year, our sales are down by not less than 60%.”

Meanwhile, Google released a paper on a new type of compression algorithm titled “TurboQuant.” The research points out that this technology can reduce the runtime memory footprint of key-value cache (KV Cache) for large language models by at least 60%. Investors quickly priced it in as follows: the shortage of AI hardware will be fundamentally alleviated, and memory demand will be reduced significantly.

The cold streak in the spot market quickly spread to the capital markets. Micron’s share price has retreated by more than 24% from its recent high, and Western Digital has also slid nearly 21% from its $777.60 peak. At the same time, last week the market value of the U.S. memory chip sector evaporated by nearly $100 billion.

Faced with the price plunge and share price rout, market participants are sharply divided on the outlook for the memory industry. Some investors believe the traditional “pig cycle” for memory has already peaked, while HSBC argues the market is overreacting. It says the market is currently at the midpoint of an AI-driven memory supercycle, with strong demand for high-end products like HBM, and the memory shortage could last for one to two years.

Buyers say “no”: Will the traditional “pig cycle” repeat?

For traders who follow traditional cycles, the market’s broad selloff is not that simple. Former reporter and well-known Taiwan-based semiconductor analyst Dan Nystedt pointed out that many bulls attribute the recent plunge to Google’s paper, but that’s only the surface. Dan believes the real reason is that prices for memory chips in some smartphones have stopped rising.

“The real reason is simpler: the prices of certain smartphone memory chips have stopped rising. Buyers ultimately said ‘no’—this is the first top signal that experienced memory-cycle investors look for before selling.”

Dan Nystedt said. Because DRAM and NAND prices are too high, some smartphone makers plan to reduce or even cancel production of mid- to low-end phones in 2026. He revealed that just two weeks ago, a buyer refused a higher DDR4 price.

Dan Nystedt compares the memory industry to a “pig cycle” in agriculture: high prices prompt companies to expand production, but building new plants takes time, and when new capacity is released at the same time, prices crash. He believes investors following that script have already exited quickly, and Micron’s and SanDisk’s share prices have therefore pulled back sharply.

Over the past 50 years, memory chipmakers have gone through a dozen major boom/bust cycles. Only since 2010 there have been three: the 3G/4G and cloud computing boom in 2012-2015; the 5G and cloud service provider expansion in 2016-2019; and a surge in PCs/servers driven by the pandemic from 2020-2023. And what began in 2024 is an upward cycle driven by AI servers (HBM and SRAM).

“Whenever someone writes ‘this time is different,’ that’s usually the classic sign that bullish sentiment has gone crazy.” Nystedt quoted a legendary trader Jesse Livermore’s saying: “The market is always right, while opinions are often wrong.” He reminds investors that when chip buyers stop panic buying, when rebounds repeatedly run into persistent selling, seasoned capital will retreat quickly according to the script.

Structural change: Are memory companies no longer “cycle stocks”?

However, regarding Dan Nystedt’s analysis, independent analyst Jukan has a different view.

He pointed out that buyer resistance to price increases is mainly focused on traditional memory like DDR4, not the entire memory market. The abnormal surge in DDR4 prices earlier was partly attributable to stockpiling by the China market, which gave smartphone makers room to adjust specs for lower-end devices.

“But DDR5 is a completely different story,” Jukan said. Smartphone and PC makers, in this year’s first quarter and even the second quarter, have been accepting the sharp DDR5 price increases honestly and straightforwardly. In today’s AI and high-end device ecosystem, DDR5 is not a bargaining chip that buyers can negotiate on; it’s a core input that must be secured even if it means paying a premium. Flagship products built around DDR5 fundamentally cannot reduce specifications.

Second, the market completely ignores the fundamental transformation in the business model of the memory giants. Jukan scoffed at the so-called “seasoned investors” who believe in “blindly selling off when spot prices fall.”

“The way memory companies operate is no longer the kind of blind capacity expansion it used to be,” Jukan said sharply. Samsung, SK hynix, and Micron—these three giants—are moving toward TSMC’s business model: after securing core customers’ advance payments (Advance payments) and visibility into long-term demand, they then build capacity.

In recent coverage, South Korean media reported that Samsung is discussing cooperation agreements based on advance payments with major players such as Microsoft. Memory giants understand better than anyone the pain that excess capacity brings to a cycle. That’s why they are now pursuing extremely restrained capacity expansion, rather than reflexive overbuilding.

Wall Street banks back it up: The memory supercycle is only at the midpoint; the market has five major concerns about overreaction

Unlike the panic sentiment in the spot market, investment banks still have strong confidence in the memory industry’s long-term prospects. In a research note released on March 30, HSBC explicitly said, “In our view, current concerns have been amplified too much; we are at the midpoint of an AI-driven supercycle.”

And the market’s current concerns are all excessive reactions. The bank listed five specific concerns:

1）Negative impact from raw material and electricity price increases caused by the Middle East conflict;

2）Slowing memory price growth in the second half of 2026;

3）Industry technologies that reduce memory usage in AI systems, such as Google’s “TurboQuant” and Nvidia’s “KVTC”;

4）Gradually increasing capital expenditure plans by major memory manufacturers;

5）Intensifying competition from Chinese memory manufacturers.

The report noted that the Middle East conflict has no substantive impact on memory manufacturers’ procurement of raw materials. Meanwhile, the impact of absolute growth in profitability on share prices will be far greater than the effect of slowing DRAM price growth rates. At the same time, memory manufacturers remain highly clear-headed and restrained in the execution of capital expenditure.

Regarding the TurboQuant technology from Google that triggered the market selloff, the bank believes it is still too early to worry. Commercialization of the technology will require about a year, and its reference parameter scale is smaller than the current AI environment. More importantly, the bank said TurboQuant alleviates memory bandwidth bottlenecks, improving system efficiency and reducing token costs, thereby accelerating the commercialization and adoption of AI. The report wrote:

“The net impact is that we believe improvements in efficiency will accelerate AI development—this is a positive development and should trigger a sharp rise in AI adoption.”

At the same time, the bank expects that 2026 AI server shipments will surge 28% year over year. From 2026 to 2027, the average DRAM content per server will achieve a strong 17% growth. With the explosive growth in AI inference demand, enterprise solid-state drives (eSSD) are entering a golden age. The report expects that by 2027, the share of eSSD in total NAND demand will jump from 18% in 2023 to 40%, and AI servers will consume 62% of that.

The bank believes the market is currently at the midpoint of an AI-driven supercycle, whose scale is comparable to the six-year DRAM shortage triggered by office automation in 1990-1995. Looking back, from 1990 to 1995, as Windows 3.0 and subsequent operating systems spread, office automation led to a structural DRAM shortage lasting six years. It drove the DRAM market size to surge from $7 billion in 1990 to 6x, to $41 billion in 1995.

The bank believes that today, infrastructure buildout driven by large models, Agentic AI, and physical AI (such as autonomous driving) will cause memory shortages to last at least one to two years.

Based on these judgments, the report strongly believes in the certainty of the benefits they will bring within the memory supercycle. Regarding the recent plunge, the report wrote: “We believe any pullback provides additional buying opportunities.”

		Sina Statement: This message is republished from a Sina partner media outlet. Sina.com publishes this article for the purpose of transmitting more information, and does not mean that Sina endorses its views or verifies the described content. The content is for reference only and does not constitute investment advice. Investors act at their own risk.

Massive information and precise interpretation—available in the Sina Finance app

责任编辑：凌辰