# HBM demand — X 热门讨论 (2026-09-16 15:01 UTC)

## @FISAYOMi67 (💫FISAYOMI) · 09-16 12:40 · ♥53 ↻5 💬44 Whether the current memory/storage chip rally is powered by real AI compute demand or just speculative hype is easily the biggest debate in market circles right now.

Instead of relying on emotional hot takes, running cyclical supply-demand dynamics and memory pricing models through Ling-3.0-flash-Fin brings some much-needed objectivity to the table.

Separating fundamental HBM/NAND demand drivers from a potential AI bubble setup is how you actually protect your downside. > 引用 @Re7_AI: 当所有人在问存储还能涨多少的时候，Ling-3.0-flash-Fin 模型给了我最准确的答案 https://x.com/FISAYOMi67/status/2100203147943608596

## @EhrmantrautCap_ (Ehrmantraut Capital) · 09-16 10:58 · ♥70 ↻3 💬5 Absolutely crazy.

Citi is expecting the DRAM shortage to become even worse in 2028, and expects undersupply to continue through 2031.

DRAM supply-demand inbalance is expected to worsen from -8.7% in 2027 to -9.7% in 2028.

NAND supply-demand inbalance is expected to peak in 2027 with -6.1% and become slightly less worse in 2028 with -5.5%.

HBM bit demand is expected to surge, growing 62% YoY in 2027 and 69% YoY in 2028.

You won’t see an end to the memory shortage any time soon.

$MU $SNDK $SKHY $KXIAY $DRAM > 引用 @pequityresearch: Citi: Memory

Citi expects memory undersupply conditions to extend through 2031.

HBM & DRAM Projections

Surging HBM Demand: HBM bit demand is projected to jump 62% YoY in 2027 (to 75.2B gigabits) and 69% YoY in 2028 (to 127.0B gigabits).

Global DRAM Growth: Global DRAM demand is forecast to grow 30% in 2027 and 35% in 2028, driven by simultaneous requirements for server DDR5 and HBM.

Severe Market Undersupply: Limited DRAM supply growth (19% in 2027; 22% in 2028) will result in tight supply-demand deficits of -8.7% in 2027 and -9.7% in 2028.

NAND & eSSD Projections

Expanding eSSD Needs: Storing massive amounts of previously learned data will drive heavy demand for enterprise SSDs.

NAND Deficits: NAND demand growth will outpace supply—projected at 29% vs. 21% in 2027 and 33% vs. 25% in 2028—resulting in supply deficits of -6.1% in 2027 and -5.5% in 2028.

Capacity Trade-Off: The supply shortage in NAND is largely driven by memory manufacturers prioritizing and expanding greenfield capacity for DRAM and HBM at NAND's expense.

$MU $SNDK $SKHY $EWY $DRAM https://x.com/EhrmantrautCap_/status/2100177376168591699

## @MelvinInvests (Melvin) · 09-16 14:23 · ♥36 ↻10 💬6 HOLY SMOKES, AI is about to create a memory shortage that could last through 2031 (Save this).

Citi expects memory shortages to persist through 2031 as AI infrastructure consumes growing quantities of high bandwidth memory, server DRAM and enterprise storage.

HBM bit demand is projected to increase 62% in 2027 and another 69% in 2028, while total DRAM demand is expected to grow 30% and 35% during those years.

The problem is that DRAM supply is projected to grow only 19% in 2027 and 22% in 2028, leaving market deficits of 8.7% and 9.7%, respectively.

A deficit does not mean that memory completely disappears but it does mean that customers collectively want substantially more memory than manufacturers can supply at current production levels.

HBM is intensifying the shortage because it requires advanced DRAM dies, complex packaging and more manufacturing capacity than conventional server memory.

As manufacturers allocate additional resources to HBM and server DRAM, less capacity remains available for other products, which spreads pricing pressure across the broader memory market.

The same imbalance is developing in NAND, where demand is projected to grow 29% in 2027 and 33% in 2028, compared with supply growth of only 21% and 25%.

AI servers need enterprise SSDs to store model weights, training data, retrieval databases and the growing volumes of information generated by AI agents.

This means the AI boom is increasing demand for both the memory located beside the accelerator and the storage systems feeding data into the cluster.

The memory manufacturers could experience several years of higher prices, stronger margins and greater revenue visibility instead of another brief commodity cycle.

Micron’s 2026 HBM production is already sold out, while additional fabrication capacity from its Tongluo facility is not expected to provide meaningful output until the second half of 2027.

SK Hynix has similarly reported that customer demand exceeds available supply, while Samsung expects AI infrastructure spending and agentic AI to keep HBM, server DRAM and enterprise SSD markets undersupplied.

Micron and SK Hynix provide the most direct exposure to HBM and server DRAM, while Samsung offers broader exposure across memory, foundry and advanced packaging.

SanDisk provides greater exposure to the NAND and enterprise SSD shortage, while the South Korea ETF offers a more diversified way to gain exposure to Samsung and SK Hynix.

Bullish on memory and if these shortages really persist through 2031, I think the market is still underestimating how much pricing power memory companies could have.

Follow @MelvinInvests for more AI memory insights, and if you want to see what stocks I'm personally buying, you can join Milk Road Pro for just $1 using the link below.

https://t.co/MMkdjfqLsf https://x.com/MelvinInvests/status/2100229197969133692