Memory Shortage 2027: Will It Really Last a Decade?
SK hynix says 2027 will be the worst supply year ever for memory and ADATA expects a decade-long shortage. Here is how long the RAM squeeze may last.
By Capital & Compute
For a year the standard question about the memory shortage was when it ends. This month the two people best placed to answer stopped talking about an end. SK hynix now calls 2027 the worst supply year the memory industry has ever seen, and ADATA’s chairman thinks the shortage runs for another decade. So the honest answer to how long the RAM squeeze lasts is no longer “into 2027.” It is “nobody selling you memory expects it to be over by 2027, and one of them just said 2035.”
How long will the memory shortage last?
On current forecasts, the memory shortage lasts at least through 2027 and probably well beyond. SK hynix expects 2027 to be the worst supply year on record and demand to outrun its capacity past 2030. ADATA is more extreme, putting the shortage at roughly ten more years. No major supplier is currently forecasting relief before late 2027, and several see the imbalance running deep into the decade.
That is a real shift from where the conversation sat even a month ago. The 2026 memory shortage explainer landed on late 2027, with parts of the industry pointing to 2028, as the credible window for the first relief. The new statements from SK hynix and ADATA do not contradict that so much as extend it: the front edge of relief still sits around 2027, but the people making the chips are now describing a shortage that gets tighter first and lingers far longer than a single-year fix.
What SK hynix actually said
The headline came from an interview SK hynix CEO Kwak Noh-jung gave to Reuters in mid-July 2026, reported by Tom’s Hardware and aggregated by VideoCardz. Two things stood out.
First, the framing of 2027. Kwak described it as the most difficult year in the history of the memory industry from a supply perspective. Not a tight year. The worst one ever recorded. Read that against the fact that DRAM has been in shortage since late 2025 and the mainstream chip price has already roughly tripled, and it says the peak of the pain is still ahead of us.
2027 will be the most difficult year in the history of the memory industry from a supply perspective.
Second, and easy to miss, the horizon. Kwak said customer demand is expected to remain above SK hynix’s production capacity beyond 2030. That is the number that should reset expectations. It is one thing for a hardware vendor to warn about next year. It is another for the world’s leading HBM maker to tell investors it does not expect supply to catch demand for the rest of the decade. This is a corporate forecast, not a guarantee, and SK hynix has an obvious interest in a market that stays tight. But it is a striking thing to say out loud.
ADATA’s decade call
If SK hynix reset the ceiling, ADATA blew through it. Chairman Chen Li-bai told reporters the DRAM shortage could last another ten years, as covered by Tom’s Hardware. His argument: the industry has badly underestimated how much memory AI infrastructure will need, and the new fabs scheduled to come online between 2028 and 2035 may still fail to satisfy demand when they arrive.
Chen is not a neutral party. ADATA sells memory modules, and a chairman forecasting a ten-year shortage is forecasting ten good years for his own margins. He also waved off worries about an AI bubble, suggesting the question should not be taken seriously until 2040 or even 2050, which is the kind of line that ages badly if the capex cycle turns. Treat the decade call as the aggressive end of the range, not the base case.
But strip out the bravado and the underlying claim is the same one SK hynix is making in calmer language. New capacity takes years to build and qualify. Almost all of it is being pointed at AI. And demand keeps revising upward faster than the fabs can be poured. Two very different executives, with very different incentives, arrived at the same shape.
The forecasts, on one timeline
Line the outlooks up on a single axis and the disagreement is narrower than the headlines suggest. Everyone agrees the shortage is real and running now. They differ on how far out it stretches.
| Item | Outlook reaches | Shortage underway (2026) |
|---|---|---|
| Gartner (relief begins) | 2027 | 2026 |
| Counterpoint (earliest inflection) | 2027 | 2026 |
| Intel CEO (no relief until) | 2028 | 2026 |
| SK hynix (demand over supply beyond) | 2030 | 2026 |
| ADATA (shortage could run to) | 2035 | 2026 |
Gartner and Intel’s CEO anchor the near edge, seeing the first relief around late 2027 into 2028. Counterpoint Research has pointed to Q4 2027 as the earliest realistic inflection. SK hynix and ADATA sit at the far end. The gap between “relief starts in 2027” and “shortage lasts to 2035” is the whole debate, and where you land in that range decides whether you buy memory now or wait.
Why HBM keeps consumer memory short
The reason none of these forecasts bends toward a quick fix is the same reason the shortage started: high-bandwidth memory. HBM is the stacked, high-margin DRAM that sits next to AI accelerators, and it eats the same wafers that would otherwise become the DDR5 in a laptop.
The allocation math is getting worse, not better. TrendForce estimates HBM will consume roughly 30% of the combined DRAM wafer input at Samsung, SK hynix, and Micron by the end of 2027, up from about 22% in 2026. Here is the twist that makes it hurt: because HBM stacks many dies and yields fewer usable bits per wafer, that 30% of wafers turns into only about 13% of the actual DRAM bits shipped. Nearly a third of the industry’s wafer capacity, producing a thin slice of the bits that go into everything else.
That is the structural trap. A wafer moved to HBM removes far more than one wafer’s worth of consumer DRAM from the market, and every forecast above assumes AI demand keeps pulling wafers in that direction. It is the same argument I made in why RAM is so expensive in 2026: the shortage is a choice about where wafers earn the most, and right now they earn the most stacked next to a GPU. The demand behind that choice is not slowing. Frontier training runs keep getting larger, a trend our breakdown of what it costs to train AI models in 2026 traces in dollars.
What it means if you buy memory
For a buyer, the practical read barely changes, it just gets more durable. If you needed RAM last month, the advice was buy what you need and skip the speculation. The SK hynix and ADATA statements only harden that. Waiting for relief is now a bet against the two firms with the clearest view of supply, and one of them just told you it does not expect balance before 2030.
ADATA’s chairman expects prices to keep rising through the back half of this year. Even his own scenario for some softening lands no earlier than the second half of 2027, and he was blunt that prices are unlikely to return to pre-2025 levels. So the realistic best case is not cheap memory. It is memory that stops climbing.
If you are trying to spend less without betting on the market, the tactics in how to save money on RAM in 2026 still apply, and the week-to-week direction is on the DRAM price tracker. For the near-term buy-now-or-wait call by buyer type, the 2026 RAM price forecast breaks it down.
The bottom line
Two executives with opposite incentives to exaggerate, the supplier who wants a tight market and the module maker who profits from one, both landed on the same picture in the same week: a shortage that peaks in 2027 and does not resolve on any near horizon. The exact end date is unknowable and the ten-year call is almost certainly too long. But the useful takeaway is not the year. It is that the era of asking “when do RAM prices come back” is being replaced by a harder question: what do you build, and buy, in a world where memory stays scarce by design.
Frequently asked questions
- How long will the memory shortage last?
- On current forecasts, at least through 2027 and likely beyond. SK hynix expects 2027 to be the worst supply year on record and demand to exceed its capacity past 2030. ADATA is more aggressive, forecasting a shortage that could last another ten years. No major supplier currently sees relief before late 2027.
- Why does SK hynix think 2027 will be the worst year?
- CEO Kwak Noh-jung told Reuters in July 2026 that customer demand keeps rising while SK hynix is limited by production capacity, making 2027 the most difficult supply year in the industry history. The core cause is AI: high-bandwidth memory for accelerators consumes wafers that would otherwise make consumer DRAM.
- Will RAM prices drop in 2027?
- Probably not by much. ADATA expects prices to keep rising through late 2026 and sees only limited softening no earlier than the second half of 2027, and says prices are unlikely to return to pre-2025 levels. SK hynix expects demand to outrun supply beyond 2030, so a sharp price drop in 2027 looks unlikely.
- Is a ten-year memory shortage realistic?
- It is the aggressive end of the range. ADATA chairman Chen Li-bai bases it on AI demand outpacing new fabs due between 2028 and 2035, but he also sells memory and profits from scarcity. DRAM is cyclical and has defied confident long-range forecasts before, so treat a decade-long shortage as a scenario, not a base case.
- Why does HBM cause the consumer memory shortage?
- High-bandwidth memory uses the same wafers as consumer DRAM but earns far more per wafer, so makers prioritize it. TrendForce expects HBM to take about 30% of the top three makers combined DRAM wafer input by the end of 2027, up from 22% in 2026, while producing only about 13% of the bits, which starves the mainstream market.
Sources
- Reuters (2026), via Tom’s Hardware. SK Hynix says 2027 will be the worst year for memory shortage, forecasts crunch to last until 2030. Secondary coverage of a Reuters interview with CEO Kwak Noh-jung; 2027 worst-year framing, demand above capacity beyond 2030. Verified 2026-07-23. tomshardware.com
- VideoCardz (2026). SK hynix expects worst-ever memory shortage in 2027, ADATA expects a decade-long shortage. Secondary aggregation of the SK hynix and ADATA statements. Verified 2026-07-23. videocardz.com
- Tom’s Hardware (2026). Adata chairman says DRAM shortage will last another 10 years, dismisses AI bubble talk until 2040 or 2050. Secondary; ADATA chairman Chen Li-bai ten-year forecast, 2028-2035 fab timing. Verified 2026-07-23. tomshardware.com
- TrendForce (2026). Tight DRAM Supply Gives Suppliers Greater Pricing Power in HBM, with HBM Contract Prices Expected to Surge Multiples Higher in 2027 (press release, June 2 2026). Independent; HBM ~30% of DRAM wafer input by end 2027 (up from ~22% in 2026), ~13% of bits. Verified 2026-07-23. trendforce.com
- TechSpot (2026). NAND shortages could finally ease in 2027, but ADATA warns the DRAM crisis may last another 10 years. Independent; 2026 NAND supply deficit 4-5%, servers 40%+ of NAND consumption. Verified 2026-07-23. techspot.com
- Counterpoint Research (2026). Advanced Memory Prices Likely to Double as DRAM Crunch Spreads on NVIDIA Pivot. Independent; Q4 2027 earliest inflection. Verified 2026-07-23. counterpointresearch.com