Memory Maker Micron Sees Shortages Worsening Through 2028


TL;DR

  • Supply Squeeze: Memory maker Micron expects shortages to worsen in 2027 and 2028, raising supply concerns for data-center and device manufacturers.
  • AI Memory: Most of its 2027 high-bandwidth memory for AI chips is contracted at significantly higher prices than in 2026.
  • Factory Ramps: New chip factories need several quarters after first output to add meaningful supply.
  • Record Revenue: Fiscal 2026 fourth-quarter revenue reached $54.23 billion, up 379% from a year earlier.

Micron expects memory shortages to deepen through 2028 because growing AI demand is absorbing capacity faster than new factories can add it. For data-center operators and device makers, the chipmaker’s September 30, 2026 earnings update points to tighter competition for supply even as industry shipments increase.

Chief Executive Sanjay Mehrotra expects demand to exceed supply in both 2027 and 2028, with no clear date for a return to balance. Micron’s fiscal 2026 fourth-quarter results, covering the period ended September 3, put quarterly revenue at $54.23 billion, against $11.32 billion a year earlier.

AI Demand Strains Memory and Storage

AI systems need several kinds of memory and storage. Dynamic random-access memory, or DRAM, holds the active data processors use. High-bandwidth memory, or HBM, is a specialized form of DRAM that gives AI accelerators rapid access to data close to the processor. NAND flash retains data in solid-state drives, or SSDs, which feed those systems with datasets and stored model information.

Larger AI models, longer conversations and more simultaneous requests increase requirements across that hierarchy, according to Micron. The pressure therefore reaches ordinary server working memory and storage as well as the stacked memory beside AI chips. Data-center SSD revenue approached $10 billion in the quarter, more than ten times the year-earlier level and over two-thirds of Micron’s total NAND revenue. Micron sees additional storage demand from moving AI context data into SSDs and replacing hard drives.

Making more HBM also consumes manufacturing capacity that could otherwise support conventional DRAM. Mehrotra’s earnings-call explanation links newer HBM generations’ greater silicon requirements with smaller productivity gains from future manufacturing processes. More investment and more advanced production technology can add output, but the company expects demand to absorb those gains.