Probably for several years, but not indefinitely—and no forecast makes that certain. Micron expects memory demand to exceed supply in 2027 and 2028, with tighter conditions than in 2026, according to an October 2026 report by Tom’s Hardware. That is a company outlook, not a guarantee about the whole industry. AI is increasing demand for several kinds of memory and storage, while new capacity takes years to build and ramp. But memory remains a cyclical business: demand, product mix and capacity can change, and a shortage can eventually give way to oversupply.
Will AI demand for memory chips last?
AI-related memory demand looks likely to remain a meaningful source of pressure into 2027 and 2028. Micron CEO Sanjay Mehrotra said the company expected demand to exceed supply in both calendar years, with tighter conditions than in 2026, as reported by Tom’s Hardware in October 2026. Micron also forecast industry DRAM and NAND shipments growing in the low-to-mid-20% range over the following two years, while still anticipating supply constraints. Those are company forecasts, not independent measurements or promises that prices will keep rising.
Micron’s SEC filing describes AI-driven data-center growth pushing demand for memory and storage beyond the company’s and industry’s ability to increase supply quickly. It says the company made supply-allocation decisions and saw customers engage more actively in longer-term commitments. This is useful evidence of what one major supplier was experiencing, but it does not establish that every memory product or supplier faces the same shortage. Micron’s 2026 Form 10-K, filed May 28, 2026, provides that company account.
For readers asking how long a memory shortage will last, the responsible answer is: Micron’s published outlook reaches through 2028, but the duration beyond that is uncertain. It depends on how quickly AI infrastructure spending and customer orders grow, which products are needed, and when announced manufacturing capacity becomes usable output.
#1 Best Overall
Which memory and storage products are affected?
“Memory chips” is not one interchangeable market. AI data centers use several product types, and tight supply in one does not automatically mean identical conditions in another.
| Product | What it does | Why it matters to AI supply |
|---|---|---|
| HBM | High-bandwidth memory is a form of DRAM built as stacked layers, designed to deliver data close to processors. | AI accelerators rely on HBM for high data throughput. Micron describes its 3D-stacked design as using through-silicon vias and offering greater bandwidth at lower power than other memory types. |
| Server DRAM | Volatile memory used by servers for active workloads. | AI systems need server memory in addition to accelerator-attached HBM. Server and PC DRAM are not necessarily the same specifications or supply pool. |
| NAND and enterprise SSDs | NAND is non-volatile storage; SSDs use it to retain data without power. | AI data centers also need large storage systems. Demand for these products is related to, but distinct from, demand for DRAM. |
Micron reported progress across HBM, server DRAM and data-center SSDs in its Q3 2026 earnings release dated June 24, 2026. Its filing also distinguishes product categories and end markets; its compute and networking business unit spans multiple products, so its revenue should not be treated as HBM-only sales. Micron’s fiscal 2025 Form 10-K gives those business and end-market distinctions.
Rank #2
- Ideal for Gifting
- Ideal for a bookworm
- Compact for travelling
Can AI demand cause a DRAM shortage beyond HBM?
It can contribute to tighter conventional DRAM supply, but the effect is not uniform. Manufacturers have finite resources and may prioritize products such as HBM that serve data-center customers. Industry reporting on Samsung and SK hynix describes concern that allocating more resources and capacity to high-margin HBM could tighten conventional DRAM as well. Tom’s Hardware’s April 2026 report attributes shortage commentary to the companies through secondary reporting; it should not be read as a direct filing or proof that every DRAM segment is short.
Client PCs, mobile devices, servers and AI accelerators also have different requirements and demand patterns. Therefore, an AI-driven shortage forecast does not by itself establish the availability or price outlook for every PC or phone memory product. The same caution applies to NAND: storage demand is part of the AI infrastructure buildout, but DRAM and NAND are different markets.
Why can’t new capacity end the shortage immediately?
Building fabs and ramping production takes time. Announced projects, new product generations and actual high-volume output are separate milestones. A facility can be planned or under construction well before it contributes meaningful supply, and a new generation does not automatically mean enough total industry output to relieve tightness.
Micron’s disclosed product milestones
In June 2026, Micron said it was shipping HBM4 in high volume to a lead customer and expected volume production of HBM4E in calendar 2027. The same release reported progress on server DRAM and data-center SSD products. These are Micron’s milestones and expectations, not a schedule for all manufacturers or evidence that aggregate supply will immediately catch up.
Rank #4
Micron’s fab investment schedule
Micron estimated 2026 capital expenditure above $25 billion, net of government incentives. In its February 26, 2026 Form 10-Q, the company projected first DRAM wafer output from its first Idaho fab in mid-calendar 2027, expected a second Idaho fab to be operational by the end of 2028, and said its first New York fab would provide supply in 2030 and beyond. These dates describe Micron’s plans, not guaranteed delivery dates or industry-wide capacity. The filing is available in Micron’s 2026 Form 10-Q.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Does this mean the memory cycle is over?
No. A prolonged shortage forecast does not prove that memory has stopped being cyclical. The sector has historically moved between constrained supply and oversupply. AI investment may extend or reshape the current cycle, but it cannot eliminate the possibility that demand slows, capacity ramps faster than expected, or customers reduce orders after securing enough inventory.
Recommended Free Tools
Best Value
- It can be a gift option
- Comes with secure packaging
- Helpful in various ways
It also matters whether demand is firmly committed or only forecast. Longer-term customer agreements and efforts to reserve supply can help suppliers plan, but they do not remove execution risk or make future demand certain. Likewise, capital spending announcements are not equivalent to production: output depends on project completion and manufacturing ramps. Micron’s outlook is most useful as a dated indication of its expectations, not as a definitive timetable for the entire market.
What should readers watch to judge how long tightness may last?
Rather than treating a single shortage headline as a market-wide verdict, separate demand, product mix and delivered capacity:
- Product: Identify whether a claim concerns HBM, conventional DRAM, NAND or SSDs. Tightness in one category does not establish the same condition in another.
- Source and scope: Distinguish a company’s outlook or filing from independent industry data. Micron’s disclosures explain its own position and plans; competitor statements cited through media reports have a different evidentiary basis.
- Demand: AI data-center expansion is a driver, but the pace of infrastructure investment and actual customer orders can change.
- Capacity stage: Separate fab plans and expected product ramps from output already shipping at scale. The dates Micron gives for Idaho and New York facilities extend from 2027 into 2030 and beyond.
- Cycle risk: Even when suppliers report constraints, memory markets can later swing toward oversupply if demand weakens or capacity catches up.
The cited sources do not provide a consistent independent series for memory prices, inventories or industry utilization, so they do not support a precise date for when shortages will end or a prediction of a particular price move.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
