Micron and Samsung executives said this week that the global memory shortage will persist for at least another two years, with Micron CEO Sanjay Mehrotra telling investors that demand for the company's memory will exceed its available supply through that period.

Mehrotra's remarks, delivered last night, refer to Micron's business-to-business sales: high-bandwidth memory (HBM) sold into AI accelerators and DRAM sold into servers. The company exited the consumer memory market earlier — Micron no longer sells consumer RAM through its Crucial brand, ending a retail business that had operated for nearly three decades.

The commercial logic behind that exit is now playing out in the wider market. Manufacturing capacity across the memory industry is being prioritized for AI and server products, where HBM and server DRAM command premium pricing. That allocation decision directly limits the supply of memory fabricated for consumer devices, tightening availability and pricing for PCs, laptops, and other end markets that depend on commodity DRAM.

For Micron, the supply-demand imbalance is not an accident of timing. HBM carries significantly higher bit pricing than standard DRAM, and AI accelerator vendors have committed to large volumes through multi-year agreements. Shifting wafer starts toward HBM and server DRAM maximizes revenue per wafer — but it comes at the cost of consumer-grade output. Mehrotra's forecast that demand will outstrip supply for at least two years signals Micron expects AI-driven demand to keep absorbing incremental capacity as it comes online.

Samsung's executives echoed the same timeframe this week, indicating the constraint is industry-wide rather than specific to one supplier's allocation choices. When the two largest DRAM producers — which together with SK hynix dominate global memory output — both describe a multi-year shortage, the effect reaches every downstream buyer.

For OEMs and device makers, the practical consequence is straightforward. Consumer DRAM supply will remain tight through roughly 2028, according to the executives' guidance, with pricing power sitting with the memory makers as long as AI demand holds. System builders sourcing commodity memory face continued cost pressure and potential allocation risk, since suppliers have little incentive to rebalance capacity toward lower-margin consumer parts while HBM and server DRAM remain sold out.

The executives' two-year outlook frames the shortage as a structural feature of the AI buildout rather than a short cyclical squeeze. Unless AI memory demand softens or new capacity arrives faster than planned, consumer device makers should expect constrained DRAM supply and firm pricing through 2028.