Micron Technology (NASDAQ:MU) expects memory-market supply constraints to extend beyond 2027, driven by continuing demand from artificial intelligence (AI), according to Sumit Sadana, Executive Vice President and Chief Business Officer, at the KeyBanc Capital Markets’ Annual Technology Leadership Forum.
The Chief Business Officer noted that customer demand has escalated since the company’s latest earnings report, with the memory market projected to be even tighter in 2027 than 2026. Micron has yet to see clear signs of when supply will meet demand.
The main bottleneck, according to Sadana, is DRAM rather than power, space, or logic wafers, largely due to the complexity and build time required for new memory fabrication plants. The rapid growth of generative and agentic AI, along with future artificial general intelligence, is reshaping the market. Agentic AI workloads require five to thirty times more memory tokens than basic chat-interface tasks—adding tremendous pressure on memory production.
High-bandwidth memory (HBM) has become critical, as some data-center customers struggle to acquire even half of the memory volumes they seek despite high pricing.
Micron is addressing market conditions with strategic customer agreements (SCAs) that include binding commitments and take-or-pay clauses through 2030. Sixteen such agreements totaling $22 billion in commitments have been secured, providing price bands aimed at sustaining margins above previous cycle highs. These agreements foster deeper collaboration on product and research roadmaps, such as Micron’s HBM3E, which boasts 30% lower power consumption than its nearest competitor.
On the investment front, Micron is boosting planned U.S. outlays to $250 billion, alongside ongoing investments in Japan, Taiwan, Singapore, and India. The company is advancing U.S. manufacturing, with its Idaho 1 facility expecting to open in mid-2027 and further expansions planned in Idaho, New York, and Virginia. U.S.-made memory is targeted to command a pricing premium and enhance supply-chain resilience.
Looking forward, Sadana expects HBM4E and “physical AI”—such as robotics and humanoid robots—to become significant demand drivers later this decade, potentially requiring vast DRAM and SSD resources.
