Micron Posts Record $54.2B Quarter, CEO Says Memory Will Be Much Tighter in 2027 and 2028
Micron Technology reported fiscal fourth-quarter 2026 revenue of $54.2 billion on September 30 — a quarterly record for the company — and CEO Sanjay Mehrotra used the earnings call to issue a supply warning that spans the next two years.
“We expect memory and storage supply-demand conditions to be much tighter in calendar 2027 and 2028 than they were in 2026,” Mehrotra said in prepared remarks.
Why Supply Gets Tighter, Not Looser
The counterintuitive forecast — that conditions tighten even as Micron, Samsung, and SK Hynix all invest in new cleanroom capacity — reflects a structural dynamic: AI data center build-out is consuming memory faster than fabs can scale.
HBM (high bandwidth memory) is the critical constraint. Every Nvidia GB200 and GB300 accelerator requires HBM3E, and Micron, SK Hynix, and Samsung have sold through their 2026 and 2027 HBM allocations to hyperscalers and compute hosters. New cleanroom space takes 18-24 months to come online, and conversion from standard DRAM production to HBM is not reversible on short timescales.
Mehrotra’s statement aligns with analyst consensus heading into 2027: IDC, Omdia, and SemiAnalysis have all modeled persistent HBM shortage through at least 2028, with spot pricing for HBM3E remaining elevated well above contract prices.
What $54.2 Billion Means
Micron’s Q4 2026 record was driven almost entirely by AI infrastructure demand, with data center memory accounting for the majority of the revenue mix. The company has progressively shifted its DRAM production mix toward HBM at the expense of consumer DRAM and standard server DDR5, where oversupply remains.
The record quarter follows a year of exceptional performance. All three major memory producers — Micron, SK Hynix, and Samsung — crossed $1 trillion in market capitalization in mid-2026 in the same month, a sector first. Micron’s Q3 2026 gross margins hit 84.6%, driven by HBM’s premium pricing.
Implications for AI Infrastructure
Mehrotra’s supply warning has direct implications for AI infrastructure planning. The hyperscalers — Google, Microsoft, Amazon, Meta — have locked in multi-year memory contracts with all three major producers. Smaller compute hosters and emerging AI clusters like SpaceXAI’s Colossus 2 are competing for the residual allocation.
For model training, memory bandwidth is often the binding constraint on how fast a cluster can train frontier models. A sustained supply tightening in 2027-28 means that even well-capitalized labs with access to GPU compute may face limits on memory throughput — slowing training iteration speed or forcing architecture choices that trade memory for compute efficiency.
Micron guided current-quarter revenue above analyst targets. Full guidance numbers were not disclosed in the preliminary reporting available at publication time.