Memory prices are still painfully high, but China's fast-expanding DRAM and NAND supply is starting to change the market's next move.
The memory cycle has moved from shortage panic to supply anxiety in a matter of weeks. Chinese chipmakers are pushing harder into DRAM and NAND, and the first visible signs are now appearing in mainstream products, just as AI data centers have been swallowing the industry's available output.
That matters because memory has become one of the quiet cost centers behind the AI boom. Graphics processors get the attention, but training clusters, inference servers, enterprise SSDs and high-memory edge devices all depend on a steady supply of DRAM and NAND. When prices spike, every data center build becomes more expensive. When they fall, the economics of AI can change quickly.
The latest signal came from reports that Corsair Vengeance DDR5 modules sold in China have been spotted using DRAM from ChangXin Memory Technologies, better known as CXMT. The module in question was described as a 16GB DDR5-6000 stick, which is not exotic by high-end desktop standards, but that is exactly the point. Chinese memory is not just a domestic policy story anymore. It is starting to show up in ordinary supply chains.
CXMT is China's most important DRAM player, while Yangtze Memory Technologies, or YMTC, remains the country's best-known NAND flash maker. Both companies have spent years operating under the shadow of U.S. export controls, but the market pressure created by AI demand has given them a clearer opening. If global buyers cannot get enough supply from Samsung, SK Hynix and Micron, they will look harder at alternative sources.
According to TrendForce, conventional DRAM contract prices were expected to rise 58% to 63% quarter over quarter in the second quarter of 2026, while NAND flash contract prices were expected to climb 70% to 75%. That followed an even sharper first quarter, when conventional DRAM was projected to jump 90% to 95% and NAND flash 55% to 60%.
Those figures explain why a new source of supply has suddenly become so important. The industry has not been dealing with a normal cyclical shortage. AI infrastructure demand has pulled capacity toward high-bandwidth memory, server DRAM and enterprise SSDs, leaving consumer devices, PCs and smaller hardware makers competing for what is left.
Western and Korean suppliers have enjoyed that pricing power, but they also know how quickly memory markets can turn. When too much capacity arrives, prices do not drift lower politely. They can fall hard, because DRAM and NAND are commodity-like markets with high fixed costs and brutal inventory swings.
The price relief may not arrive tomorrow
The near-term picture is still tight. TrendForce has said meaningful enterprise SSD capacity expansion is unlikely until late 2027 or 2028, and former Samsung semiconductor chief Kye-hyun Kyung recently warned that a surge in Chinese capacity could begin to alter the market from the second half of 2027 or the first half of 2028. That is not immediate relief for buyers ordering servers this quarter.
But markets price direction before they price arrival. If OEMs believe Chinese supply will become more dependable, they gain leverage in contract talks. If module makers believe more wafers are coming, they become less desperate to stockpile at any price. And if Samsung, SK Hynix and Micron see a future glut forming, they may respond by slowing commodity output rather than defending every unit of share.
This is where the U.S.-China chip war becomes more complicated. Washington has tried to slow China's progress in advanced semiconductors, especially the chips needed for AI. Memory is different. DRAM and NAND do not always require the same leading-edge logic processes, and China has a strong incentive to build domestic supply because every AI server, smartphone, electric vehicle and industrial device needs memory somewhere in the stack.
For AI startups, cheaper memory would be more than a procurement win. It could lower the cost of training runs, make inference clusters easier to expand and bring larger local models into reach for devices that cannot afford premium configurations today. A mid-market hardware company that can ship more DRAM in a laptop, workstation or edge box suddenly has more room to support on-device AI features without pushing prices out of reach.
There is a risk on the other side. A sharp correction would hit memory margins, pressure chip equipment orders and remind investors that AI infrastructure is still tied to old-fashioned component cycles. The same supply chain that looks permanently constrained during a boom can look overbuilt once new fabs ramp and buyers finish stockpiling.
The next thing to watch is not whether memory prices fall this month. They probably will not in any meaningful way. The better question is whether Chinese supply becomes trusted enough, and large enough, to change contract behavior before the new capacity fully arrives. If that happens, the AI buildout gets a little cheaper, and the memory supercycle starts to look less permanent than it did just a few months ago.
Also read: Kevin Warsh takes the Fed as Bitcoin waits for the next rate signal • Nomura scraps 2026 Fed rate cut forecast amid sticky inflation • Minnesota's prediction market ban is forcing a new legal fight
