The AI Memory Shortage: Why DRAM Supply Will Stay Tight Through 2028

What Is Causing the AI-Driven Memory Shortage

The shortage is fundamentally an AI demand shock colliding with structural supply constraints. The core drivers are:

Explosive HBM demand from AI accelerators. Nvidia, AMD, and Google GPUs require enormous quantities of HBM. Micron's revenue nearly quadrupled year-over-year in fiscal Q3 2026, almost entirely from HBM sales . The company's entire HBM capacity for 2026 was sold out before the year even began .

HBM cannibalizes standard DRAM capacity. HBM3E consumes roughly three times the wafer area of standard DDR5, and yield losses during vertical stacking amplify the wafer demand further . Shifting fab lines to HBM directly reduces the supply of commodity DRAM used in PCs and servers.

Agentic AI and inference workloads. Beyond training, emerging "agentic AI" workloads — autonomous AI agents running continuous inference — are creating a new, persistent layer of DRAM demand that did not exist in prior cycles .

Supply cannot keep up. Micron has repeatedly stated it can meet only half to two-thirds of customer demand from key clients . The gap is not a short-term mismatch. New greenfield fab construction takes 3–5 years. Micron's first new U.S. fab wafer output is not expected until mid-2027 , and competitors' capacity arrives even later .

How Long Micron Expects the Shortage to Last

Micron has progressively lengthened its outlook as demand has accelerated:

• December 2025: "Tight past 2026"
• June 2026: "Persist beyond calendar 2027"
• August 2026: 2027 supply conditions will be even tighter than 2026; industry supply "substantially short of demand for the foreseeable future"

The shortage is now viewed as structural, not cyclical, because meaningful new industry supply does not arrive until 2028 at the earliest .

Wall Street analysts agree. Morningstar sees supply tightness persisting well into 2027, and J.P. Morgan also expects the shortage to last through 2027 . A Kearney PERLab analysis published in 2026 projected the shortage could last at least until 2030 .

Micron's $250 Billion U.S. Investment Plan and Capacity Timeline

In response, Micron has announced the most aggressive domestic manufacturing expansion in U.S. semiconductor history. The company now plans to invest more than $250 billion through 2035 , up from a prior $200 billion commitment .

Micron also has simultaneous expansion projects in Japan, Singapore, India, and Malaysia, with new global capacity beginning to come online from 2027 onward .

Capital expenditure is ramping dramatically: FY2026 spending is raised to roughly $25 billion (from $13.8 billion in FY2025), and 2027 capex is expected to be even higher .

Strategic Customer Agreements: $100 Billion Locked In

As of fiscal Q3 2026, Micron has signed 16 Strategic Customer Agreements (SCAs) . These are not ordinary supply contracts — they are legally binding take-or-pay agreements that fundamentally transform Micron's business model:

• Coverage: Roughly 20% of Micron's DRAM volume and one-third of its NAND volume over the contract period .
• Total contracted value: Approximately $100 billion in minimum guaranteed revenue, with $22 billion in cash deposits and upfront commitments .
• Terms: Data center and consumer SCAs are typically 5-year terms (2026–2030); automotive SCAs are 3-year terms .
• Automotive partners: Signed July 2026 with Qualcomm, Visteon, HARMAN, JOYNEXT, DENSO, Astemo, and Hyundai Mobis . Additional agreements with General Motors and Ford were announced separately .
• Pricing floor: The SCAs include floor prices with "very robust gross margins" well above Micron's peak quarterly margins in any past cycle .
• Long-term target: Management expects SCAs to eventually cover roughly 50% of long-term memory revenue .

These agreements mean that even as new capacity eventually comes online, a significant portion of Micron's output — and its pricing — is locked in at premium levels through 2030.

The Broader Context

U.S. policy support. The Trump administration's push for domestic chip production is explicitly cited as a factor in Micron's accelerated U.S. investment plans . CHIPS Act funding is also supporting the buildout .

Competitor timelines matter. SK Hynix committed about $38 billion to two new fabs in South Korea. The first of those, a DRAM plant in Yongin, does not open a clean room until December 2028 . This means the entire industry's new capacity arrives on a similar delayed schedule.

What remains unconfirmed in the sources. The available evidence did not cover several other potential developments — including Section 232 tariffs on memory imports, the Pax Silica pact, Apple DRAM-related product delays, a U.S. price-fixing lawsuit against memory makers, or Nvidia spec changes affecting HBM demand. These topics may be covered elsewhere but were not verifiable from the documents collected for this analysis.

The Bottom Line

The AI-driven memory shortage is not a temporary squeeze. It is a structural reallocation of the world's semiconductor wafer capacity toward high-margin AI memory products, and it will take years to resolve. Micron's own timeline suggests 2028 before supply and demand begin to meaningfully rebalance — and that assumes no further acceleration in AI deployment.

For anyone buying a PC, server, or data center GPU in the next 18–24 months, the message is clear: memory will remain expensive and hard to get.