Memory Stocks Face a New AI Reality Check
Memory chip stocks are soaring on AI demand, but investors warn the cycle can turn fast if supply rises or model efficiency improves.

Memory chip stocks are soaring on AI demand, but investors warn the cycle can turn fast.
Memory chips have powered one of the hottest trades in global markets, with Samsung Electronics, SK Hynix, Micron Technology, and SanDisk all posting huge gains in 2026. The problem, as several fund managers told CNBC, is that the memory business has a long history of sharp booms followed by painful busts.
That tension matters now because AI demand has pushed high-bandwidth memory, or HBM, into a tight supply regime. But the same AI wave that lifted pricing can also shrink memory needs if model makers keep improving efficiency.
| Company | 2026 stock move | What it means |
|---|---|---|
| Samsung Electronics | 114% | One of the biggest beneficiaries of HBM demand |
| SK Hynix | 186% | Market leader in memory exposure |
| Micron Technology | 141% | U.S. memory maker riding the same cycle |
| SanDisk | 156% | Another fast mover in the memory trade |
| Google TurboQuant | 6x less memory | Compression method aimed at lowering LLM memory use |
The rally is real, and so is the old pattern
Get the latest AI news in your inbox
Weekly picks of model releases, tools, and deep dives — no spam, unsubscribe anytime.
No spam. Unsubscribe at any time.
The rally in memory stocks started with the launch of ChatGPT in December 2022, which drove a surge in demand for HBM chips used in training and running large language models. The market has treated that demand as durable enough to justify much higher valuations across the sector.

Investors who have lived through past memory cycles are less relaxed. Their point is simple: demand can spike, supply can catch up, and pricing can fall much faster than bullish forecasts assume.
- Samsung Electronics is up 114% year-to-date in 2026.
- SK Hynix is up 186% year-to-date in 2026.
- Micron Technology is up 141% in 2026.
- SanDisk is up 156% in 2026.
- Samsung and SK Hynix together account for more than 50% of South Korea’s Kospi index.
Why AI may help memory and hurt it at the same time
Google introduced TurboQuant on March 24, saying the compression method could cut memory needs for large language models by six times. That matters because memory demand has been one of the biggest beneficiaries of AI infrastructure spending.
When a model needs less memory, it can reduce pressure on the very chips that have been in short supply. That is why TurboQuant hit memory shares hard when it was announced.
“In the long run it’s a pretty dreadful industry,” William de Gale, portfolio manager at BlueBox Asset Management, told CNBC. “I suspect that’s still the case every time people make an argument that the memory cycle is gone.”
Deutsche Bank also warned in a Tuesday note that investors should “continue to brace themselves for continuous AI-related disruption.” That is a useful framing: AI is not a single demand story, it is a constant source of product changes that can shift who wins and who gets squeezed.
For memory makers, that means today’s shortage is not a guarantee of tomorrow’s pricing power. If model efficiency improves faster than expected, the market could move from scarcity to excess sooner than bulls want to admit.
Supply can improve faster than the market expects
Jon Cunliffe, head of investment office at JM Finn, said production has room to rise meaningfully over the next three years, especially if AI demand grows at a more normal pace. That kind of supply response is exactly what memory investors tend to underestimate during strong upcycles.

He also warned that current share prices assume several optimistic conditions at once: high prices lasting for a long time, disciplined capital spending, and profit margins staying well above historical norms. Those assumptions can all look reasonable in a shortage and all look expensive once the cycle turns.
- William de Gale described memory as an industry with “enormous ups and downs.”
- Jon Cunliffe warned that current prices assume long-lasting high margins and disciplined investment.
- Andrew Lapping of Ranmore Fund Management said investors should be careful when an industry with average historical returns is priced for very high future returns.
Andrew Lapping put it bluntly: “A leopard does not often change its spots.” It is a useful reminder that a better narrative does not erase a sector’s operating history.
Korea is the center of gravity for this trade
The concentration risk is especially clear in South Korea, where Samsung and SK Hynix have driven the Kospi to strong gains across 2025 and 2026. When two stocks make up more than half of an index, the index can look healthier than the underlying economy.
That is why some strategists are urging investors to take profits instead of assuming the rally can run forever. Standard Chartered global chief investment officer Steve Brice said peak optimism around Korean equities may not be far off, and he told CNBC he was advising clients to rotate into globally diversified portfolios.
There is still a bullish case, of course. Nomura sees SK Hynix reaching 4 million won and Samsung Electronics climbing to 590,000 won over the next 12 months, which would imply about 20% upside for Samsung and a doubling for SK Hynix from current levels. Those targets show how much optimism is still embedded in the trade.
What investors should watch next
The key question is whether the current shortage is a true structural shift or just a strong phase in a familiar cycle. The answer will depend on two things: how fast AI labs improve memory efficiency, and how quickly chipmakers add supply.
If efficiency gains like TurboQuant spread widely while production continues to rise, memory stocks could lose some of the pricing power that investors are paying for today. If demand keeps outrunning supply, the rally can continue, but the margin for error is getting thinner.
For now, the smartest move is to treat memory stocks like cyclical stocks, because that is what they have been for years. The next big test is whether earnings can keep up with the share prices, or whether the market has already priced in too much perfection.
// Related Articles
- [IND]
OpenAI updates its Europe privacy policy
- [IND]
OpenAI is right to keep ads out of sensitive chats
- [IND]
AI bootlegs are already draining streaming royalties
- [IND]
AMD and Microsoft push Windows ML on GPU and NPU
- [IND]
OpenAI’s IPO filing turns hype into scrutiny
- [IND]
Skatteetaten proves public sector AI should be judged by outcomes