Analysts warn memory, storage glut 2027

Thomas Coughlin, an IEEE president in 2024 and a Forbes contributor covering storage markets, wrote on May 19 that the AI data center buildout could swing from shortage to oversupply by 2027 if power, water and permitting constraints delay enough projects. His argument was that memory and storage makers are expanding against aggressive data center plans, while a meaningful share of those projects may never arrive on schedule. The warning landed against a market that has spent the past year focused on tight supply and rising prices for DRAM, high-bandwidth memory and NAND flash. It also cut against other 2026 forecasts that still project a strong memory upcycle through 2027. ### Which analyst warning set off the latest glut debate? May 19 was the date of Coughlin's note, published by Forbes and echoed in a post on X referenced in the original discussion. Coughlin wrote that expanded production of solid-state memory and storage, combined with slower-than-expected AI data center construction, "could result in a glut by 2027." The warning did not argue that AI demand is fading. Coughlin said the risk comes from a mismatch between paper plans for data center expansion and the physical realities needed to complete those projects, including power generation, grid upgrades, water availability and local opposition. ### Why are power and water suddenly central to a memory story? (forbes.com) The U.S. power requirement in Coughlin's piece was 780 gigawatts of additional demand by 2030 for planned data center projects, compared with a current national peak load of 759 gigawatts. He wrote that such an expansion would be difficult to achieve. (forbes.com) Water was the second constraint. Coughlin wrote that U.S. data center direct water consumption is on track to double or quadruple 2023 levels annually, and that about two-thirds of U.S. data centers built since 2022 are in high water-stress areas. He also cited polling showing about 70% of Americans oppose data centers in their communities, with concerns including utility bills, water use, noise and air pollution. (forbes.com) ### How much of the announced data center pipeline is considered real? Brett Forster, vice president of renewables at McCarthy Building Companies, said at Latitude Media's Transition-AI 2026 conference in April that only one in four data center projects is considered real and likely to move forward, according to Coughlin's account. Latitude Media separately reported 780 gigawatts of announced U.S. data center projects and quoted Aurora Energy Research managing director Oliver Kerr calling it "an additional U.S. of power by 2030." (forbes.com) That gap between announced capacity and buildable capacity is the core of the glut thesis. If chipmakers and storage suppliers add output based on the larger pipeline while actual commissioning slips, component demand could undershoot supply in 2027. That is an inference from the reported project and capacity figures, not a separate forecast from Coughlin. (forbes.com) ### Why does this clash with the rest of the 2026 memory narrative? January 22 was the date of TrendForce's published forecast that the global memory market would reach $551.6 billion in 2026 and $842.7 billion in 2027, with 53% year-over-year growth in 2027. TrendForce said AI systems are driving structurally higher demand for DRAM and NAND, and that cloud service providers have sharply increased procurement. (forbes.com) January 26 was the date CNBC published comments from Synopsys Chief Executive Sassine Ghazi, who said the memory chip crunch would continue through 2026 and 2027. Ghazi said most memory from the top suppliers was "going directly to AI infrastructure" and that new capacity takes a minimum of two years to come online. Those forecasts describe a market still constrained by supply. (trendforce.com) Coughlin's May warning describes a different failure mode: not too little memory capacity, but too many planned bits chasing AI systems that are delayed by the grid, water and politics. ### What should readers watch next to see which view is winning? (cnbc.com) 2027 is the date both sides of the debate point to, but the nearer markers are project completion and utility interconnection progress in 2026. Latitude Media's conference reporting and Coughlin's May 19 article both point readers back to the same variables — power, water and permitting — while TrendForce's January forecast remains the clearest published benchmark for how much DRAM and NAND revenue the industry still expects by 2027. (forbes.com)