Standing beside an open car trunk on a side street in southern China’s tech hub of Shenzhen, Roy Luo runs a quick, practiced eye over the labels on a stack of NAND flash memory chips, the kind vital for phones, computers, and servers.

He has only minutes to decide. Prices can shift without warning, and high-value stock may vanish to another buyer. Luo, a 30-year-old electronics trader, checks his phone, nods once, and wires payment to the seller.

In Huaqiangbei, the world’s largest electronics market, this is how some of the most lucrative deals have been made since last September, as a global shortage of memory chips fuels more speculation.

Luo says his net profits have multiplied several times compared with a year ago. “People are making millions now,” he said. “Some have cleared 50 million yuan ($7.2 million) or more.”

The frenzy is being driven by surging demand for AI infrastructure, which requires far more memory than conventional computing. As manufacturers allocate more capacity to AI-related memory, supplies for smartphones, laptops, and other electronics have tightened worldwide, leaving prices in China’s spot market swinging sharply from day to day as buyers rush to secure stock.

Market research firm TrendForce said memory prices have been rising in recent months, with conventional DRAM — which runs applications — contract prices expected to jump roughly 90% to 95% quarter on quarter, while NAND flash, which stores data, contract prices could rise about 55% to 60%.

The impact is already reaching consumers. In late December, Chinese smartphone maker Xiaomi unveiled its flagship Xiaomi 17 Ultra, with the 16 GB plus 512 GB model priced at 7,499 yuan — about 500 yuan more than the previous generation with a comparable configuration.

Ahead of the launch, company president Lu Weibing said rising memory costs were putting direct pressure on pricing. Local rivals, including Vivo and OPPO, have also raised prices for high-end models.

With global memory production still concentrated among South Korea’s Samsung Electronics and SK Hynix, and the U.S.-based Micron Technology — and with meaningful new capacity years away — the supply gap is unlikely to close soon.

While some analysts say the shortage could give Chinese memory makers a limited window to expand output and capture market share, industry insiders see little chance of a return to stable prices or predictable supply. Instead, those inside the storage market are learning to navigate a market where scarcity, speed, and risk now shape every deal.

Big data

Luo had waited since 2023 for the memory trade to pay off. When prices stalled that year, he quit a high-paying job at tech giant Alibaba and began trading chips full time, tracking inventory and price swings for signs the market was finally turning.

By April 2025, the signals were unmistakable: more traders were entering the market, volumes were rising, and profits were widening. Rather than turning stock quickly like a traditional dealer, Luo builds inventory ahead of expected price moves and sells only after momentum takes hold. “This round drew in far more participants,” he said. “And the money was bigger.”

As prices surged, discussion of soaring chip costs spread rapidly across Chinese social media. On microblogging platform Weibo, a hashtag comparing the value of a box of memory chips to an apartment in Shanghai topped the trending charts. Users shared screenshots of rapidly changing quotes, joked that prices were rising faster than gold, and complained that waiting even a day could add thousands of yuan to the cost of a computer.

Behind the online chatter, control of memory supply was shifting upstream.

Michael Li, who leads a supply team focused on SK Hynix memory chips, said popular and high-capacity models are now repriced almost daily, with orders often dependent on immediate payment. Miss the price or delay the transfer, and the deal can be canceled the same day. “This is no longer a buyer’s market,” Li admitted. “Everything moves fast now.”

And with distributors short on inventory, Li increasingly sources chips from traders like Luo, regardless of formal authorization. “As long as the chips are new and unopened, the market accepts them,” he said. “It doesn’t matter if they were made five years ago or last year.”

But not every trader has benefited. What once required modest capital now demands far deeper reserves, said Luo, squeezing out smaller firms unable to hold inventory through sharp price swings. Many suffered losses in 2024 and the first half of 2025 and are still waiting for prices to recover. “The threshold is much higher,” he said. “Small companies are being squeezed out.”

Li has seen the consequences ripple downstream too. As prices climb, many domestic end users can no longer afford the most sought-after chips. Instead, high-visibility components with the steepest gains are increasingly sold overseas, where manufacturers often have greater purchasing power.

“Within China, a lot of trading now just circulates among intermediaries instead of flowing to end manufacturers,” Li said.

New normal

Jerry, a manager overseeing AI-related hardware solutions at a Chinese telecom company, has watched the memory shortage hit the company’s server business in real time.

Since July, the cost of a single memory module has surged from just over 1,000 yuan to more than 6,000 yuan, forcing repeated price hikes to customers and constant recalculations inside the firm.

“Locking in prices in advance — once standard industry practice — has effectively become impossible,” Jerry said, requesting use of a pseudonym because he was not authorized to speak publicly. “In today’s market, holding memory can be more profitable than buying gold.”

Government technology projects, which can take months to approve and bid, have been hit especially hard.

Jerry recalled a case last summer in which a major server manufacturer secured a contract worth several billion yuan, only to see memory prices surge soon after. Rather than deliver hardware at a loss, the company chose to absorb penalties for breaching the agreement — a decision that, he said, sent shockwaves through the industry.

Behind the turmoil is a surge in demand for AI-capable servers, which require far more memory than conventional systems, and are consuming a growing share of global supply. The shift accelerated in early 2025 as large-scale models spread rapidly following the launch of DeepSeek, the domestic large language model that drew global attention for its low training costs and open-source approach.

“Nvidia alone is absorbing massive volumes of memory,” Jerry explained, “while companies such as Google — along with a growing number of domestic GPU manufacturers — are competing for the same capacity, driving today’s price surge.”

As supplies tighten, global producers are increasingly directing limited shipments to their largest and most strategic customers, forcing smaller buyers to seek alternatives. In China, that scramble is giving domestic memory makers such as ChangXin Memory Technologies and Yangtze Memory Technologies a chance to expand their presence in a market dominated by foreign suppliers.

And though switching to domestically produced memory chips can introduce compatibility and stability risks, the severity of the shortage is leaving buyers with little room to refuse their terms. “There’s nothing we can do,” Jerry said. “At least they’re in China, which makes negotiation easier.”

ChangXin, for instance, now also requires cash on delivery, eliminating the grace periods that once allowed companies to pay months after receiving goods. “We would take delivery, sell the products, and then pay ChangXin after receiving payment from clients,” Jerry explained. “That’s no longer possible. Without upfront payment, there’s no supply.”

While negotiating a contract worth 20 million yuan with a client between July and December, Jerry said prices rose three to four times, with single increases of 1 million to 2 million yuan. The client initially struggled to understand the swings but ultimately accepted the terms after comparing prices across the market.

An industry insider at a leading domestic chipmaker, who asked to be identified by the pseudonym Zack because he was not authorized to speak publicly, said the volatility is already reshaping strategies across the sector.

Some customers who once planned to build their own computing centers are turning instead to cloud services or leased computing power, while others are scaling back ambitions and starting with smaller models. “Regardless of strategy,” Zack said, “there is no way to bypass the hard constraint of storage costs.”

The surge in storage prices, insiders say, suggests a return to cheap memory and loose supply is unlikely.

“As artificial intelligence becomes infrastructure, memory and GPU memory are increasingly treated as foundational resources — akin to electricity or broadband — and their prices are likely to remain elevated for years to come,” Zack said.

Editor: Apurva.

(Header image: VCG)