Two seemingly opposite views are circulating in the semiconductor industry today.
On one side, Sanjay Mehrotra, CEO of Micron, argues that memory supply may not be able to keep up with AI-driven demand for years to come. The explosive growth of AI infrastructure, HBM, and data center investments could create a prolonged period of tight supply.
On the other side, some industry observers believe that a new memory “chicken game” could emerge around 2028. Their argument is based on the rapid growth of Chinese players such as CXMT in DRAM and YMTC in NAND, combined with aggressive capacity expansion by Samsung, SK hynix, and Micron.
Interestingly, both views may be correct.
The Current Reality: Demand Exceeds Supply
Today, the memory market is fundamentally different from the commodity DRAM market of the 1990s.
HBM is not simply a memory chip.
It requires:
- TSV integration
- Hybrid bonding
- Advanced packaging
- Customer qualification
- Thermal management
As a result, increasing supply is much more difficult than adding conventional DRAM capacity.
This is why HBM remains in tight supply despite massive investments across the industry.
From this perspective, Micron’s view appears reasonable. AI demand is real, growing, and supported by enormous investments from NVIDIA, Microsoft, Google, Amazon, and Meta.
But History Offers a Different Warning
As someone who experienced the memory industry during the 1990s, I cannot help but remember the 1995 DRAM boom.
At that time, demand also appeared unstoppable.
The industry expanded aggressively.
Companies believed that even larger fabs would still struggle to satisfy demand.
For a while, they were right.
Then supply caught up.
Soon after, DRAM prices collapsed and the industry entered one of the most painful pricing wars in its history.
The lesson is simple:
High profitability always attracts more investment.
And more investment eventually creates more supply.
Is HBM Different?
Many investors argue that HBM is fundamentally different from commodity DRAM.
They are correct, to some extent.
The technological barriers are significantly higher than in previous memory cycles.
However, one rule of capitalism never changes:
Extraordinary profits attract competitors.
Samsung, SK hynix, Micron, and emerging Chinese suppliers are all responding to today’s strong profitability.
The question is not whether supply will increase.
The question is how quickly supply will increase relative to demand.
So When Should Investors Buy Memory Stocks?
Ironically, the best time to buy memory stocks has rarely been when profits were at their highest.
Historically, the most attractive opportunities often appeared when:
- Prices were depressed
- Inventories were high
- Industry sentiment was negative
- Investors had given up on the sector
Conversely, risk often increases when:
- Everyone expects endless growth
- Capacity expansion accelerates
- Earnings reach record highs
- “This time is different” becomes the dominant narrative
Looking Toward 2028
Today, we are still in a supply-constrained environment.
The short-term outlook remains favorable for memory manufacturers.
However, investors should monitor future capacity expansion more carefully than quarterly earnings.
If Samsung, SK hynix, Micron, and Chinese suppliers simultaneously increase capacity over the next several years, the market may begin pricing in future oversupply long before it actually occurs.
Conclusion
The biggest mistake in the memory industry is not buying when earnings are weak.
It is believing that exceptional profitability will last forever.
I do not believe today’s AI-driven memory market is identical to the DRAM boom of the 1990s. HBM, advanced packaging, and AI infrastructure have fundamentally changed the industry.
However, history teaches us an important lesson:
Demand can grow faster than expected.
But supply eventually tries to catch up.
For investors, the critical question is not whether memory is currently in short supply.
The critical question is:
When will supply growth begin to exceed demand growth?
The answer to that question will likely determine the winners and losers of the next memory cycle.