Storage chip prices plummet: spot market "crashes," futures market still out of stock

The storage chip market is being driven by the investment logic of “buy the rise, not the fall,” leading to a pullback in the spot market for memory modules.

On the news front, on March 25th, Google Research Institute released an AI memory compression algorithm called TurboQuant, claiming it can compress the most memory-intensive KV Cache (key-value cache) in large model inference by at least 6 times, while achieving an 8-fold performance acceleration. The U.S. stock storage chip sector plummeted accordingly, and prices in the storage chip spot market, which have been rising since last year, have begun to decline.

However, in Shenzhen’s Huaqiangbei, this round of memory module price drops started even earlier. Since mid-March, both DDR4 and DDR5 memory modules have begun to adjust prices, with reductions of 20%-30% so far. A seller in Huaqiangbei told the Beijing News Shell Finance reporter that the reasons behind this are complex, involving upstream manufacturers’ sell-offs and downstream retail investors’ panic-driven selling.

Nevertheless, the contract market for storage chips has not been affected by the price drops. Industry analysts say that by the end of March 2026, applications such as servers, PCs, and smartphones are all facing shortages of memory, and with AI and data center construction continuing to advance, the price trend for storage remains bullish.

“Two prices in one day”

Memory module market from “super profits” bull market to panic bear market

“Last year and this year’s market are like a bull and bear market.” Duan Tianting, who runs a memory module business in Huaqiangbei, told the Beijing News Shell Finance reporter that fluctuations are normal market phenomena, but the large swings in storage chip prices over the past two years are truly rare.

The spot wholesale market for storage chips in Huaqiangbei is currently experiencing a “two prices in one day” decline. Several merchants in Huaqiangbei said that since mid-March, memory module prices have fallen by 20%-30%, with multiple price changes occurring daily.

Regarding the source of this price drop, a common explanation within the Huaqiangbei memory community is that in mid-March, Samsung and SK Hynix began clearing inventory. The earliest upstream distributors received news and conducted two large-scale sell-offs on the 13th and 14th, leading to a short-term oversupply situation, with pressure transmitted to larger merchants.

Behind this decline is also a complex psychological game between buyers and sellers. Having experienced significant price swings before, Duan Tianting analyzed that this reflects a “buy the rise, not the fall” mentality. When prices rise, everyone chases orders; once prices start falling, downstream customers become cautious. Many sellers initially can hold out—“because they benefited from some gains last year”—but when prices fall below their psychological bottom line, panic selling ensues. “The more this happens, the fewer people are willing to buy.”

Several industry practitioners told Shell Finance that retail demand for storage chips is somewhat suppressed by prices. For example, individual consumers assembling computers tend to wait and see, hesitant to purchase at current prices.

Despite the ongoing decline in the storage chip market, some still want to enter the industry.

“I’m outside, debating whether to get in,” said a member of a storage chip discussion group. In just over five hours, more than 150 people joined, including veteran traders and newcomers observing the market. Screenshots shared by the group owner show multiple chat groups named after memory, storage chips, and electronic components, all full.

Duan Tianting observed that since last year, many “outsiders” have started doing memory module business, “thinking they can make money.” In 2025, storage chips are entering a price increase cycle, with a single memory module rising hundreds or thousands of yuan, expanding profit margins for traders. But Duan believes that “super profits” are not the norm. Before last year, the wholesale market for memory modules was in a “micro-profit era,” earning only a couple of yuan per module.

AI applications and model sizes are expanding in tandem

Supporting long-term bullishness for storage chips

In this round of price declines, some brands and models of chips have been spared.

“There is some randomness; not all manufacturers are lowering prices. Some brands and models have dropped more,” said Wu Wei, who works in PC assembly, to Shell Finance. He noted that the prices of Biwin 48G and Acer 32G chips have fallen quickly, by 300-400 yuan. However, compared to the threefold increase in DDR4 and DDR5 prices since last year, the decline is not large.

Long-term, storage chip prices remain bullish, especially in the contract market. Many terminal manufacturers of smartphones and computers primarily work with upstream chip suppliers through contracts.

Wang Yuqi, an analyst at TrendForce, predicts that contract prices for storage chips will remain bullish through 2026, with this upward trend expected to continue until the end of the year. Although recent market fluctuations are short-term, ongoing AI and data center development provide support for storage prices, and short-term volatility does not alter the long-term growth trend of the industry.

The supply-demand imbalance supporting rising storage chip prices still exists. According to analyst Xu Jiayuan of TrendForce, by the end of March 2026, shortages of memory in servers, PCs, and smartphones will persist. The overall shortage of DRAM (dynamic random-access memory) will only be alleviated after new capacity from manufacturers comes online around 2028.

Regarding Google’s research results that caused a “shake” in many storage chip giants’ stock prices, Wang Yuqi believes that TurboQuant’s compression of KV Cache, while reducing memory usage without affecting model accuracy, is fundamentally about improving resource utilization rather than reducing overall demand. As inference costs decrease, AI application scenarios and model sizes tend to expand simultaneously, which in the long run will continue to drive overall storage demand growth.

(Per the interviewee’s request, Wu Wei is a pseudonym.)

Beijing News Shell Finance Reporter: Zhang Xiaohui | Editor: Chen Li | Proofreader: Mu Xiangtong