Micron and the memory market: analysts expect strong AI-driven demand through the end of the decade

The memory market, which for decades has lived by a harsh cycle of overproduction and shortage, may be entering a longer phase of sustained demand. Melius Research analysts believe generative AI is changing the very structure of DRAM and NAND consumption, which means the familiar pattern for manufacturers — sharp growth, price collapse, losses — may no longer work the way it used to. Investors heard the signal: expectations of long-lasting demand lifted interest in Micron and Sandisk shares.

Historically, memory has been one of the most volatile segments of the semiconductor market. Manufacturers spent years expanding capacity, then faced oversupply, falling prices and pressure on margins. The whole sector knows this cycle well: a slowdown in demand for smartphones, PCs or servers was enough to flip the balance quickly. That is why any talk of a more stable trajectory in memory was previously met with caution.

But generative AI has added a new type of demand to the industry, one that is not tied solely to consumer electronics refresh cycles. It stems from the multi-year build-out of computing infrastructure, where memory is bought not as an auxiliary component but as a critical part of the whole system.

The reason is that modern AI systems need markedly more memory at every level of the infrastructure. Training large models requires accelerators with high-bandwidth memory, large volumes of DRAM in servers, and fast storage for data, checkpoints and working sets. Serving models in production does not remove the need either: the more requests a system handles, the longer the context and the more complex the multimodal scenarios, the greater the load on memory and the storage attached to it. That makes memory not a secondary component but one of the key limits on the performance and cost of AI infrastructure.

In other words, demand is growing not only because of the number of chips, but because each AI server needs a more expensive and denser memory configuration than systems of previous generations.

Against this backdrop, the Melius Research thesis looks like a bet not on another short recovery cycle but on a structural shift lasting to the end of the decade. If capital spending by cloud platforms, model developers and enterprise data centers stays high, memory demand will be supported for several years in a row. This matters especially for Micron, because the market increasingly values the company not merely as a player in the traditional DRAM cycle but as a supplier that benefits from the expansion of AI infrastructure.

The positive reaction to Sandisk shares shows that investors are extending this logic to data storage, where fast flash solutions are also becoming critical for AI workloads. Under this logic, even a modest improvement in the supply-demand balance can trigger a stronger revaluation of companies than in an ordinary market recovery.

There is another important point: AI demand is spread more widely than it may seem at first glance. Initially the drivers were the largest cloud providers and foundation model developers, but enterprise customers are now joining in, building their own clusters, launching internal assistants and upgrading storage for new types of workloads. This does not cancel the dominant role of hyperscalers, but it makes the overall picture less dependent on a single segment. The more AI use cases move from experiments to permanent operation, the more durable memory demand becomes overall.

Still, it would be premature to talk about the complete disappearance of cyclicality. Competition among the major players has not gone anywhere, and any overly rapid expansion of supply can again push prices down. Much will also depend on whether hyperscalers and large corporations sustain their current pace of AI investment. But the framing of the question has already changed: the market is now discussing not just the next quarter and price recovery, but how deeply AI is rewriting the fundamental demand profile in memory.

The main takeaway is that for memory makers, generative AI may become not a temporary spike but a longer-lasting pillar for the business. If this scenario is confirmed, the industry gets a rare chance to break out of its familiar model of sharp swings and move to more predictable growth. For investors the signal is simple: Micron, Sandisk and the entire memory segment are increasingly viewed as direct beneficiaries of the multi-year expansion of AI infrastructure, not just participants in another technology cycle.