Global Memory: Updating our DRAM D-S model; Industry imbalance set to persist

研报英文原文证据摘录

Global Memory: Updating our DRAM D-S model; Industry imbalance set to persist

Deutsche Bank

Research

North America Industry Date

United States 16 June 2026

Global Memory

TMT Industry Update

Semiconductors

Updating our DRAM D-S model;

Industry imbalance set to persist

Melissa Weathers

With this note, we are updating our industry DRAM Demand-Supply (D-S) model to

Research Analyst

incorporate the latest DRAM demand data (1Q26) and to update our supply-side +1-212-250-2134

wafer start per month (WSPM) forecasts. At a high level, our updated estimates

suggest that current DRAM demand-supply tightness could persist well into 2028 Apoorva Kumar

Research Associate

and potentially beyond as we believe DRAM bit supply growth will still lag very

+1-212-250-9286

strong DRAM bit demand growth. This marks a more constrained environment

relative to our last update, which called for a shortage at least through 2027. We Ross Seymore

expect the next leg of tightness to stem from growing demand for traditional and Research Analyst

low-power DRAM, as agentic AI drives AI workloads to be even more memory- +1-415-617-3268

intensive (see takes from our recent MU NDR here). Notably, we see these tight

supply conditions persisting even despite new & accelerated capacity additions

announced in the last 180 days. For reference, we discuss key changes to our D-S

model below:

n Tailwinds from Agentic AI— We believe the rise of agentic AI (LLMs that Key Changes

can think and act on their own) will drive more CPU-based workloads in the Company Target Price Rating

coming years, and we believe this will significantly boost the need for 1,000.00 to

MU.OQ

conventional and low-power DRAM chips. For an AI to be truly considered

-

agentic, it must be able to act autonomously (within predefined 1,500.00