The memory cycle may have already peaked. AI demand can keep growing and these stocks can still go lower.

Micron, SK hynix and SanDisk are getting beaten up despite reporting record numbers. MU is 42% below its high, SK hynix 35%, and SNDK 63.5%. Meanwhile, the S&P 500 is barely 4% off its record. That’s quite a difference when the memory companies are supposed to be among the biggest winners of the AI boom.

I wonder if the market is already looking past the shortage.

TrendForce’s September report still expects DRAM prices to rise 10–15% in Q4 and NAND prices 15–20%. But buried in the same report are some less encouraging details. PC makers are cutting SSD capacity in mainstream computers to save money. Buyers have inventory. NAND wafer prices are getting too expensive for some downstream customers, and module makers are starting to get cautious.

In other words, the people buying memory are feeling the price increases. They can’t just keep paying more forever. Record pricing is great for Micron and SK hynix, but eventually customers start looking for ways around it.

China isn’t helping the long-term picture either. CXMT reported progress on a denser memory chip design in October. It doesn’t have to compete with Micron and SK hynix on equal terms tomorrow. Just the possibility of more supply down the road can change how investors value these companies.

And AI itself is getting cheaper. Epoch AI’s September research found that the cost of achieving a given level of AI performance has fallen about 47% per quarter over the past three years. That is an incredible pace. More efficient models and inference techniques mean companies can do more with the hardware they already have.

Maybe that leads to so much additional AI usage that hardware demand keeps exploding. That’s certainly possible. But I wouldn’t automatically assume every dollar of additional AI spending translates into higher memory prices and fatter margins.

SK hynix just reported an 80% operating margin. Revenue reached 104.6 trillion won, up 328% year over year. It has long-term agreements covering more than half of its 2027 and 2028 production capacity, and management expects shortages to continue through 2030. Yet the stock is down 35% from its high.

At some point, investors have to decide whether these earnings are sustainable or whether they’re buying companies near the top of a very profitable cycle. Those are two different things.

Think about Cisco after the dot-com bubble. The internet kept growing. Networking equipment remained essential. Cisco didn’t suddenly become a bad company. But buying it near the peak was still a terrible investment for a long time.

I don’t know if memory prices have peaked. But I wouldn’t be surprised if these stocks have already seen their highs, even with AI spending continuing to grow. You don’t need an AI crash to get a memory stock crash. You just need today’s extraordinary profits to become tomorrow’s ordinary profits.

Not financial advice