Micron and Memory Market: Analysts Expect Strong AI-Driven Demand Through Decade's End

The memory market, which for decades has lived by a strict cycle of overproduction and shortage, may be entering a longer phase of sustained demand. Melius Research analysts believe that generative AI is changing the very structure of DRAM and NAND consumption, which means the scheme familiar to manufacturers—"sharp growth → price collapse → losses"—may no longer work as it did before. Investors heard this signal: on expectations of prolonged demand, interest in Micron and Sandisk stocks has grown.

Historically, memory has been one of the most volatile segments of the semiconductor market. Manufacturers would spend years expanding capacity, then face oversupply, price declines, and margin pressure. Such a cycle was well known throughout the sector: it was enough for demand for smartphones, PCs, or servers to slow down, and the balance would quickly shift the other way.

This is why any talk of a more stable trajectory in memory was previously viewed with caution. But generative AI has introduced to the industry a new type of demand that is not tied solely to the refresh cycle of consumer electronics. This concerns the multi-year construction of compute infrastructure, where memory is purchased not as a supporting component, but as a critical part of the entire system.

The reason is that modern AI systems require significantly more memory at all levels of the infrastructure. Training large models requires accelerators with high memory throughput, large volumes of DRAM in servers, and fast storage for data, checkpoints, and working sets. The need does not disappear when deploying models in production: the more requests the system handles, the longer the context, and the more complex the multimodal scenarios, the higher the load on memory and associated storage.

This makes memory not a secondary component, but one of the key constraints on AI infrastructure performance and cost. In other words, demand is growing not only because of the number of chips, but also because each AI server requires 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 the next short recovery cycle, but on a structural shift through the end of the decade.

If capital expenditures by cloud platforms, model developers, and enterprise data centers remain high, demand for memory will be supported for several years in a row. For Micron, this is especially important because the market is increasingly evaluating the company not simply as a participant in the traditional DRAM cycle, but as a supplier benefiting from the expansion of AI infrastructure. The positive reaction to Sandisk shares shows that investors are extending this logic to the data storage segment as well, where fast flash solutions are also becoming critical for AI workloads.

In such logic, even moderate improvement in the balance of supply and demand can lead to stronger revaluation of companies than in a typical market recovery phase. There is also another important point: AI demand is distributed more widely than may appear at first glance. At first, the drivers were the largest cloud providers and developers of fundamental models, but now corporate customers are increasingly entering the game, building their own clusters, launching internal assistants, and modernizing storage for new types of workloads.

This does not negate the dominant role of hyperscalers, but makes the overall picture less dependent on a single segment. The more AI use cases transition from experiments to continuous operation, the more sustainable the demand for memory becomes as a whole. At the same time, it would be premature to speak of the complete disappearance of cyclicality.

Competition among major players has not gone anywhere, and any too rapid expansion of supply can again put pressure on prices. Moreover, much will depend on whether the current pace of AI investment by hyperscalers and large corporations continues. But the very formulation of the question has already changed: now the market is discussing not only the near-term quarter and price recovery, but how deeply AI is rewriting the fundamental profile of memory demand.

The main conclusion is that for memory manufacturers, generative AI can become not a temporary spike, but a more lasting foundation for business. If this scenario plays out, the industry will have a rare chance to move away from the familiar model of sharp swings and transition to more predictable growth. And for investors, the signal is clear: Micron, Sandisk, and the entire memory segment are increasingly being viewed as direct beneficiaries of multi-year expansion of AI infrastructure, rather than merely as participants in yet another technology cycle.