Memory prices spike
DRAM and NAND flash costs are forecast to more than double this year, and analysts say supply disruption could persist into 2028, tightening capacity for memory-dependent services. Industry commentary also reports TSMC’s 2nm and 3nm wafer slots are fully sold out into 2027, concentrating advanced-foundry risk in a small set of providers. (ciodive.com) (seekingalpha.com)
Memory chips are getting sharply more expensive, and analysts now say the squeeze on supply could last for years. (ciodive.com) Gartner said on April 13 that dynamic random-access memory and NAND flash costs are on track to more than double in 2026, pushing total semiconductor spending above $1.3 trillion. The firm said the disruption could run into 2028. (ciodive.com) TrendForce said on February 2 that first-quarter 2026 contract prices for conventional dynamic random-access memory were revised up to 90% to 95% quarter over quarter, while NAND flash was revised up to 55% to 60%. It said cloud service providers, server makers and personal computer makers were all facing supply gaps. (trendforce.com) Memory is the short-term workspace for servers and phones, while NAND flash is the long-term storage used in solid-state drives and devices. When both rise at once, hardware makers pay more to build servers, laptops and storage systems. (trendforce.com) The pressure is coming from artificial intelligence buildouts that consume large amounts of advanced memory and from suppliers shifting capacity toward higher-margin products. TrendForce said on March 31 that major suppliers were continuing to pull back from consumer dynamic random-access memory, leaving shortages unresolved. (trendforce.com) The strain does not stop at memory. Commentary cited by investors and industry trackers says Taiwan Semiconductor Manufacturing Co.’s most advanced 3-nanometer and 2-nanometer wafer capacity is effectively booked out well into 2027 as major chip designers line up for leading-edge production. (seekingalpha.com) (trendforce.com) Taiwan Semiconductor Manufacturing Co. says its 3-nanometer process entered high-volume production in 2022, and its 2-nanometer family is scheduled for volume production in the second half of 2025, with an enhanced N2P version due in the second half of 2026. That means the same foundry network feeding the fastest artificial intelligence chips is also ramping through a narrow set of factories. (tsmc.com) (investor.tsmc.com) TrendForce reported in January 2025 that Taiwan Semiconductor Manufacturing Co.’s Baoshan and Kaohsiung 2-nanometer fabs were each expected to reach 60,000 to 65,000 wafers a month by late 2026 or early 2027. In August 2025, it also cited reports that a 2-nanometer wafer could cost as much as $30,000, about 50% above 3-nanometer pricing. (trendforce.com 1) (trendforce.com 2) Computer makers are already reacting. CIO Dive reported last month that HP took “targeted pricing actions” after memory chip prices doubled during the three months ended January 31, and Gartner expects data-center capital spending to hit $1 trillion in 2026. (ciodive.com) If the forecasts hold, buyers of servers, storage and artificial intelligence infrastructure will be competing for pricier memory and scarce leading-edge manufacturing at the same time. Gartner’s timeline puts that pressure not in one quarter, but through 2028. (ciodive.com)
