Micron and the memory market: industry players don't expect prices to fall noticeably for several more years

Memory prices, which used to follow a familiar cycle of rises and falls, may stay high for several more years. Market players warn that the boom in artificial intelligence systems is reshaping demand so strongly that the old logic of falling prices no longer looks guaranteed.

Why the cycle is changing

The memory market lived in waves for decades. Manufacturers ramped up output, prices rose amid shortages, then supply caught up with demand and a period of decline began. This model suited both producers and buyers: companies could wait out the peak and buy later at lower prices.

Now that scenario looks less reliable. Where smartphones, PCs and consumer electronics were once the main driver, more and more demand now comes from data centers and the infrastructure for training and running AI models.

The problem for buyers is that the new demand did not simply add to the old. It changed the priorities of the whole industry. Production lines, investment and engineering resources increasingly target the types of memory needed for accelerators, servers and cloud platforms. That lowers the odds of rapid overproduction, and with it the usual price crash after a peak. The market looks less and less like past cycles, when falling prices seemed a matter of time.

What is holding prices up

Market players point to several factors that make an early price decline hard to count on:

• AI servers need noticeably more memory per node than ordinary enterprise systems.
• Production capacity expands slowly, and bringing new lines online takes years.
• Higher-margin segments pull suppliers away from cheap mass-market products.
• Large cloud companies are ready to buy up big volumes even at high prices.

In this configuration, the market depends less on weakness in individual electronics categories. Even if demand from PCs or smartphones temporarily sags, the corporate sector may offset it. For manufacturers, that means steadier revenue. For customers, it means little hope of a quick return to the price levels considered normal a few years ago. In effect, expensive memory is ceasing to be a brief anomaly and becoming the baseline planning scenario.

Consequences for the industry

A long period of expensive memory changes the economics not only for chipmakers but for the entire supply chain. Server platforms, accelerators, storage systems and ready-made cloud services all get pricier. Companies building AI infrastructure have to reserve budgets in advance and count the cost of each configuration more carefully.

This is especially painful for startups and teams with limited resources: higher memory prices quickly turn into higher costs for training models, inference and scaling products.

There is another effect. If the market gets used to memory selling at a higher price, the very notion of a normal price level shifts. That is why industry players talk not only about several years of expensive memory but about a more unpleasant scenario: no guaranteed price decline at all. In other words, this is no longer about temporary overheating but about a possible structural shift, in which memory becomes a strategically important and less accessible component of the AI era.

What this means

For the AI market, this is bad news on cost and good news for memory makers. Buyers, especially in the server and cloud segments, will have to plan purchases on the assumption that cheap memory may simply not be available in the coming years. That means not only components but also the services built on top of this infrastructure will get more expensive.