Memory and component tightness grows

Memory is the working data layer inside every device, and chip buyers say that layer is getting harder to secure as artificial intelligence demand pulls supply toward premium parts. (electronics.org) A Global Electronics Association report released April 13 said 62% of manufacturers now report constrained availability or longer lead times for memory, while 82% report higher prices and only 14% expect improvement within six months. The group said demand for high bandwidth memory, the stacked memory used with artificial intelligence accelerators, is diverting capacity from conventional dynamic random-access memory and NAND flash. (electronics.org) That squeeze is showing up in market pricing. DigiTimes reported April 14 that Nanya Technology’s first-quarter profit reached NT$26 billion as dynamic random-access memory prices rose more than 70%, and separately reported that Silicon Motion President Wallace Kou said NAND flash prices have climbed 4 to 10 times since August 2025. (digitimes.com) Memory stores data close to the processor so a system can keep working without waiting on slower storage. High bandwidth memory does the same job at much higher speed by stacking chips vertically next to artificial intelligence processors, which makes it more valuable and gives suppliers a reason to favor those products. (electronics.org, news.samsung.com) The result is not limited to cloud servers. The association said the shift is affecting smartphones, laptops, vehicles, industrial equipment, and medical devices, where memory is a basic bill-of-materials item and longer lead times can delay production plans. (electronics.org) The bottleneck is also spreading beyond memory chips themselves. DigiTimes said April 13 that shortages are extending into multilayer ceramic capacitors, the tiny components that smooth and store electrical charge on circuit boards, as artificial intelligence hardware demand pulls more parts into the same supply chain. (digitimes.com) At the leading edge, foundry capacity remains tight too. TSMC’s investor site says its 2 nanometer N2 process entered volume production in the fourth quarter of 2025 and that the follow-on N2P process is scheduled for volume production in the second half of 2026, while analysts cited by Seeking Alpha say 3 nanometer and 2 nanometer capacity is effectively sold out into 2027. (tsmc.com, seekingalpha.com) Suppliers are trying to add output rather than walk away from the market. Samsung said in February that its high bandwidth memory sales should more than triple in 2026 from 2025 levels and that it is expanding high bandwidth memory 4 production capacity, but those expansion plans still point to a market where the fastest memory is being reserved first. (news.samsung.com) For electronics companies outside the artificial intelligence buildout, the near-term problem is basic procurement. The Global Electronics Association said 94% of companies can still source memory, but most are doing so with more limits, higher costs, and less predictable delivery than they faced a year ago. (electronics.org)