RAM prices plummet, memory stocks tumble collectively—has the memory supercycle peaked?
TechFlow Selected TechFlow Selected
RAM prices plummet, memory stocks tumble collectively—has the memory supercycle peaked?
HSBC believes that current market concerns are overblown, and the AI-driven memory supercycle is only at its midpoint, with shortages likely to persist for one to two years.
Navigating Web3 tides with focused insightsAuthor: Long Yue
Recently, memory prices—which had been rising steadily for several months—suddenly reversed course, sparking market concerns that the memory cycle may have peaked.
According to market-tracking data, DDR5 memory modules sold by multiple U.S. retailers have experienced widespread price cuts, with some kits dropping as much as $100. Take Corsair’s VENGEANCE series, for example: its 32GB kit operating at up to 6400MHz is currently priced at approximately $379.99—down sharply from a recent peak of $490, representing a single-kit decline of over $110.
The domestic market has likewise been hit hard. A wholesaler told China Business News that mainstream 16GB memory module prices “plummeted over RMB 100 in a single day,” prompting large-scale liquidation by distributors who had previously stockpiled inventory.
“Prices collapsed outright starting last Saturday,” admitted Mr. Wang, a long-time memory equipment wholesaler at Beijing’s Bai Nao Hui electronics market. He showed the extreme price trajectory of a mainstream 16GB 3200MHz memory module: it cost just over RMB 130 in May last year, surged to a peak of RMB 980 in December, and—after months of volatility at elevated levels—has now fallen back to around RMB 700.
Mr. Wang lamented that soaring prices had exhausted consumer demand expectations: “Only those with urgent needs are buying. Compared to before November last year, our sales volume has dropped by more than 60%.”
Meanwhile, Google published a paper on a novel compression algorithm named “TurboQuant.” The research states this technology can reduce the memory footprint of key-value (KV) caches during large language model (LLM) inference by at least 60%. Investors quickly priced in the implication: AI hardware shortages will be fundamentally alleviated, and memory demand will shrink substantially.
The chill from the spot market rapidly spread to capital markets. Micron Technology’s share price has retreated over 24% from its recent high, while Western Digital fell nearly 21% from its peak of $777.60. In addition, the U.S. memory chip sector lost nearly $100 billion in market capitalization last week.
Faced with plunging prices and collapsing stock valuations, market participants have grown sharply divided over the memory industry’s outlook. Some investors believe the traditional memory “hog cycle” has peaked; HSBC, however, argues that market concerns are overblown—that we are currently midway through an AI-driven memory supercycle, with robust demand for high-end products such as HBM, and memory shortages likely to persist for one to two years.
Buyers Say “No”: Is the Traditional “Hog Cycle” Repeating?
For traders steeped in traditional cycles, the market plunge isn’t so straightforward. Dan Nystedt, former journalist and renowned semiconductor analyst based in Taiwan, points out that many bulls blame Google’s paper for the recent crash—but that’s merely a surface-level explanation. According to Nystedt, the real cause is simpler: prices for certain smartphone memory chips have stopped rising.
“The real reason is far simpler: prices for certain smartphone memory chips have stopped rising. Buyers have finally said ‘no’—and that’s the first topping signal experienced memory-cycle investors look for before selling.”
Nystedt explains that DRAM and NAND prices have become so high that some smartphone manufacturers plan to scale back or even eliminate production of mid- and low-end devices by 2026. He revealed that buyers rejected higher DDR4 pricing just two weeks ago.
Nystedt likens the memory industry to agriculture’s “hog cycle”: high prices spur producers to expand capacity, but new fabs take time to build—and when multiple new plants come online simultaneously, oversupply triggers a sharp price collapse. Investors following this script have already exited swiftly, dragging Micron’s and SanDisk’s share prices sharply lower.
Over the past 50 years, memory chips have undergone more than a dozen major boom-and-bust cycles. Since 2010 alone, there have been three: the 2012–2015 surge driven by 3G/4G rollout and cloud computing; the 2016–2019 expansion fueled by 5G deployment and cloud service providers’ growth; and the 2020–2023 PC/server spike triggered by pandemic-related demand. The current upswing, beginning in 2024, is powered by AI servers—specifically demand for HBM and SRAM.
“Whenever someone writes ‘this time is different,’ it’s often a classic sign that bullish sentiment has gone irrational,” Nystedt quotes legendary trader Jesse Livermore: “The market is always right, and opinions are usually wrong.” He cautions investors that when chip buyers stop panic-buying and rallies repeatedly meet persistent selling pressure, seasoned capital departs swiftly—following the well-rehearsed script.
Structural Transformation: Are Memory Companies No Longer “Cyclical Stocks”?
Yet independent analyst Jukan offers a counterpoint to Dan Nystedt’s analysis.
He notes that buyer resistance to price hikes is largely confined to legacy memory types like DDR4—not the entire memory market. The anomalous DDR4 price surge was partly attributable to hoarding in the Chinese market, giving smartphone makers room to adjust specs on entry-level devices.
“But DDR5 is an entirely different story,” Jukan observes. Smartphone and PC manufacturers accepted substantial DDR5 price increases in Q1 and even into Q2 this year. Within today’s AI and premium-device ecosystem, DDR5 is not a negotiable item—it’s a core input buyers must secure, even at a premium. Flagship products built around DDR5 simply cannot downgrade their specifications.
Secondly, the market has completely overlooked the fundamental transformation underway in memory giants’ business models. Jukan scoffs at so-called “seasoned investors” who dump shares blindly upon spotting spot-price declines.
“Memory companies no longer operate via reckless, indiscriminate capacity expansion,” Jukan keenly observes. The three industry leaders—Samsung, SK hynix, and Micron—are increasingly adopting TSMC’s business model: building new capacity only after securing advance payments and clear visibility into long-term customer demand.
Recent Korean media reports confirm Samsung is already negotiating prepayment-based partnerships with tech giants including Microsoft. Memory leaders understand all too well how excess capacity can destroy cyclical stability. Their current priority is extremely disciplined, restrained capacity expansion—not reflexive overbuilding.
Banks Stand Firm: The Memory Supercycle Is Only Midway—Five Market Concerns Are Overblown
In contrast to the panic in spot markets, investment banks remain confident about the memory industry’s long-term prospects. In its March 30 research report, HSBC explicitly stated, “In our view, current concerns are significantly overblown; we are squarely in the middle of an AI-driven supercycle.”
The report identifies five specific concerns currently dominating market sentiment—all of which it deems overreactions:
1) Negative impacts from Middle East conflict on raw material costs and electricity prices;
2) Slowing memory price growth in H2 2026;
3) Industry-wide technologies—including Google’s “TurboQuant” and NVIDIA’s “KVTC”—designed to reduce memory usage in AI systems;
4) Rising capital expenditure plans among leading memory manufacturers;
5) Intensifying competition from Chinese memory vendors.
The report notes that the Middle East conflict has had no material impact on memory manufacturers’ raw material procurement. Moreover, absolute profit growth will influence stock prices far more than any moderation in the DRAM price-growth slope. Meanwhile, memory manufacturers continue to execute capital expenditures with remarkable clarity and restraint.
Regarding Google’s TurboQuant—the very technology that triggered the market sell-off—the bank argues it’s premature to worry. Commercialization remains roughly one year away, and its reference parameter scale falls short of current AI environments. More importantly, the bank points out that TurboQuant alleviates memory bandwidth bottlenecks, improves system efficiency, and lowers token costs—thus accelerating AI commercialization and adoption. The report states:
“The net effect is that we believe efficiency gains will accelerate AI development—an unequivocally positive event expected to drive a sharp increase in AI adoption rates.”
The bank also forecasts AI server shipments will surge 28% year-on-year in 2026. Between 2026 and 2027, average DRAM content per server is projected to grow strongly by 17%. And with AI inference demand exploding, enterprise SSDs (eSSDs) are entering a golden era. The report projects eSSD share of total NAND demand will soar from 18% in 2023 to 40% by 2027—with AI servers consuming 62% of that volume.
The bank believes the current market is squarely in the middle of an AI-driven supercycle—one comparable in scale to the six-year DRAM shortage triggered by office automation between 1990 and 1995. Historically, from 1990 to 1995, the proliferation of Windows 3.0 and subsequent OS versions drove a six-year structural DRAM shortage, pushing DRAM market size from $7 billion in 1990 to $41 billion in 1995—a sixfold increase.
Today, infrastructure build-out spurred by large models, agentic AI, and physical AI (e.g., autonomous driving) will sustain memory shortages for at least one to two years.
Based on these assessments, the report firmly affirms the high degree of certainty that these companies will benefit from the memory supercycle. Regarding the recent plunge, it concludes: “We believe any pullback presents an additional buying opportunity.”
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
