Deutsche Bank Model: DRAM Supply-Demand Shortage May Extend Beyond 2028
Taylor Wilson
Deutsche Bank's updated supply-demand model shows the DRAM shortage could last beyond 2028, extending its prior forecast by at least a year; the core driver is an explosion in server memory demand fueled by agentic AI.
Why has the shortage timeline been pushed out again?
Deutsche Bank's previous call was "shortage through at least 2027." The new model pushes that to beyond 2028 — because demand growth is outrunning new capacity by a wide margin.
This means → DRAM bit-supply growth will lag demand growth for years to come. The industry isn't nearing relief; every model update makes the gap wider.
Deutsche Bank also refreshed both sides of the model — incorporating Q1 2026 demand data and updated monthly wafer-starts-per-month (WSPM) projections on the supply side.
How exactly does agentic AI drive DRAM demand?
The core force is agentic AI — large language models that can reason and act autonomously. Deutsche Bank expects these models to sharply lift demand for conventional DRAM and low-power DRAM over the next several years.
The key bottleneck sits at CPU servers. CPUs typically do not integrate HBM (high-bandwidth memory designed for GPUs), so conventional DRAM becomes the primary memory for AI workloads.
In plain terms = GPU servers run AI on HBM; CPU servers have no HBM and must rely on conventional DRAM. The more agentic AI workloads, the more DRAM those CPU racks consume.
Deutsche Bank points to Nvidia's CPU-only Vera rack: its projected DRAM configuration reaches 400 TB — 5 to 10 times that of Nvidia's GPU-based racks.
How much did Deutsche Bank raise its demand forecast?
Agentic AI factors are now formally built into the model: server shipments are projected at a 16% CAGR from 2025 to 2030, while DRAM content per server application is projected at a 24% CAGR over the same period.
This means → server DRAM bit demand is being amplified on two axes — more servers shipping, and each server packing more memory.
Can supply expansion close the gap?
Deutsche Bank raised its WSPM forecasts to reflect fab expansions announced in the past ~90 days: Samsung P4 adding 120K WSPM; SK Hynix M15X and Yongin Fab 1 adding 80K and 360K WSPM respectively; Micron's expansion at Powerchip adding 45K WSPM; Nanya's new fab adding 45K WSPM.
In total, Deutsche Bank now projects 1.475 million WSPM of new capacity over the next five years, up from 1.2 million previously.
In plain terms = yes, capacity is expanding — but the model still shows a shortage beyond 2028. Even with every announced expansion baked in, supply cannot catch demand.
What does this mean for Micron?
Following a recent non-deal roadshow with Micron, Deutsche Bank raised its earnings forecast: CY2027 EPS of $160, with gross margins expected to stay above 80% for the foreseeable future.
Deutsche Bank reiterated a Buy rating and raised its price target to $1,500.
This means → if the DRAM shortage cycle holds as modeled, Micron's elevated margins have structural support — not just a one- or two-quarter tailwind.
What variable should investors watch next?
The market is focused on the mechanics of Micron's previously disclosed Strategic Customer Agreements (SCAs): post-settlement pricing, whether customers share capex burden, and how large a share of Micron's business these agreements may eventually cover.
This reflects a deeper question: whether supply-demand tightness can translate into sustained pricing power, not just a capacity shortfall on paper.
In plain terms = a shortage is one thing; the ability to keep raising prices and locking in profit through that shortage is the real test of whether these earnings forecasts can be met.
Content is for reference only, not financial advice.