The Shortage Is Building Its Own Competitor

The Wall Street Journal reported Sunday that Apple has been testing memory chips from ChangXin Memory Technologies across product lines including iPhones and MacBooks, to mitigate a component shortage driven by the AI buildout. Apple held early talks with CXMT — China's largest chipmaker by market value — with the goal of using the components in some devices sold in China, according to people familiar with the matter. Apple and CXMT did not respond to requests for comment. Laptop makers HP and Acer have already started using CXMT memory chips in devices sold outside the United States to ease supply shortages. CXMT has risen to become the world's fourth-largest maker of memory chips, including the DRAM used in smartphones, laptops and servers. Reuters previously reported that CXMT is considering building a second memory plant in Beijing. US restrictions prevent Apple from ordering customized CXMT parts; standard components remain permitted. Apple is reported to be seeking White House approval.

Every memory datapoint of the past six months has pointed the same direction. Samsung says no meaningful supply increase before 2028. SK Hynix has committed $38.1 billion to two new fabs whose capacity arrives in 2029. DigiTimes reports 2027 DRAM and HBM capacity sold out across all three major suppliers, with buyers receiving only sixty to seventy percent of what they requested. Nvidia is reportedly evaluating a lower memory configuration for its next flagship accelerator because the DRAM does not exist to build it as designed. Tim Cook called memory pricing a hundred-year flood. This is what a company does about a hundred-year flood. Apple does not switch memory suppliers casually. It has spent two decades building the most conservative, most tightly qualified supply chain in consumer electronics, and its DRAM has come from Samsung and SK Hynix. Qualifying a Chinese supplier means accepting reputational risk, regulatory risk and political risk — and Apple is doing it anyway, across two product lines, six months after it first began evaluating CXMT and YMTC. The reason is not that CXMT got dramatically better. The reason is that there is nothing else available at any price. Which is the part worth sitting with. Export controls were built to slow Chinese memory by denying it equipment and denying it customers. The equipment restriction has been partially worked around with older lithography. The customer restriction is now being dismantled by scarcity — not by policy, not by Chinese diplomacy, but by Western OEMs who cannot fill orders. HP and Acer are already shipping. Apple is testing. The qualification threshold that took years to build is being crossed because the alternative is not shipping product. A supplier's hardest problem is never manufacturing. It is getting a first tier-one customer to certify the part. The shortage is handing CXMT that certification for free. The implication cuts against the memory trade rather than for it. Every incremental Western customer CXMT wins is capacity that eventually competes with Samsung, SK Hynix and Micron in conventional DRAM — the commodity tier, where the pricing power currently sits. Not in HBM, and not soon; CXMT is nowhere near the high-bandwidth tier. But the shortage thesis rests on supply staying constrained through 2028, and this is the mechanism by which that assumption erodes from a direction the models do not include. The constraint left is political, not technical. Standard parts are permitted; customized parts are not. Apple is scoping the plan to devices sold in China and seeking approval in Washington. Whether that approval is granted is now one of the more consequential decisions in the memory cycle, and it will be made by people optimizing for something other than semiconductor supply.