Lisa Su Says AI Demand Will Stay ‘Very, Very High’ for Years as AMD Races to Add Supply

Advanced Micro Devices Inc. (NASDAQ:AMD) CEO Lisa Su says demand for AI and computing capacity will stay elevated for years, forcing the chipmaker to substantially increase supply again in 2027 even after expanding capacity this year. Speaking during a one-day visit to Taiwan Tuesday, where she met several AMD suppliers, Su said the company needs more advanced wafer capacity. She also said AMD plans to increase its more than $10 billion Taiwan supply-chain investment announced in May. "There’s very, very high demand for the next several years," Su said. AMD shares rose more than 2% in premarket trading Tuesday. AMD Still Needs More Supply AMD has already increased supply through 2026. It hasn’t been enough. "The demand is even higher than our supply," Su said of CPUs. AMD will "substantially increase our supply in 2027," she said. "But we can definitely use more." The squeeze extends beyond processors. Su said memory remains supply-constrained, including the high-bandwidth memory used in AI servers, while AMD also needs more advanced packaging capacity. That echoes a memory crunch that could last years, with Micron saying more than 75% of its 2027 output is already committed. Su met Foxconn Chairman Young Liu as well as executives from Acer, Asus and Quanta before heading to Hsinchu to meet with TSMC executives and other suppliers. AMD Is Planning Five Years Ahead Asked whether AMD might invest directly in a chip fabrication plant, Su said the company plans to stick with its model of partnering with major suppliers. But AMD is planning much further ahead with suppliers. Su said the company used to plan capacity one or two years in advance. Now it is coordinating three to five years out to ensure wafer, packaging and other capacity expands together. "We’re increasing that number going forward because our demand has continued to go up," Su said of AMD’s Taiwan investment. AI Boom May Face a Different Constraint Morgan Stanley estimates U.S. data-center developers face a 32-gigawatt power shortfall through 2028, potentially shifting part of the AI bottleneck from chips toward electricity. Even with those constraints, prediction markets are assigning little probability to a sharp reversal this year. Polymarket traders see about a 6% chance of an “AI bubble burst” by Dec. 31, according to the platform’s market. AMD is still seeing demand as new systems reach customers. Su said the company began shipping its Helios AI systems in the third quarter and continues to see "strong demand for more compute" as production ramps. Image: Shutterstock Read Also: Microsoft Could Be SpaceX's Next Big AI Customer: Bernstein Sees AI Revenue Surging Nearly 500%