The message from Sanjay Mehrotra was meant to be heard beyond the chip industry. In an interview with CNBC this week, the Micron chief executive warned that the global shortage of memory chips could stretch well past 2026, because demand from artificial-intelligence data centers is growing far faster than the industry can add capacity. “In the coming years, AI data centers’ demand for memory chips will remain strong,” he said, “and supply will struggle to keep up.”
The warning comes at a moment when memory has become the most constrained part of the AI supply chain. Micron, one of the world’s three dominant memory makers alongside Samsung and SK Hynix, has been running its fabs at full tilt and still cannot satisfy its customers. Mehrotra’s public comments this week are the most direct statement yet from a memory chief executive on how long the shortage will last.
An industry reshaped by AI
The nature of the shortage has changed. In previous cycles, memory demand came from personal computers and smartphones, and supply caught up quickly because the products were standardized. AI data centers are different: they consume memory at a rate no earlier cycle matched, and they want the highest-bandwidth, highest-capacity products the industry can make. The result is a structural mismatch — a boom that looks like the old cycles but behaves differently.
Mehrotra pointed to the industry’s response, which he says has been slow because of how badly the last downturn scarred it. Memory prices collapsed in 2023, falling to a third of prior levels, and manufacturers slashed capital spending. Micron cut its annual investment from over $12 billion to roughly $7.7 billion that year. The plants that were not built then are the capacity that is missing now, and new fabrication facilities take years to construct.
The shift toward AI-specific memory
The business is also changing shape. Manufacturers are increasingly steering production toward AI-specific products — high-bandwidth memory for accelerators, and enterprise-grade DRAM for data centers — and away from consumer lines. Micron, for its part, has been reducing its consumer business and concentrating resources on higher-margin enterprise and data-center products, according to the company’s disclosures and Mehrotra’s remarks. The strategy is a bet that AI demand is durable enough to justify abandoning markets the industry once relied on.
The industry’s other leaders are signaling the same view. Samsung has discussed longer-term memory contracts, and the chairman of SK Hynix’s parent group has said the shortage could stretch toward the end of the decade. Long-term customer agreements, once rare in a business dominated by spot-market pricing, are becoming common — a sign that customers no longer expect the squeeze to resolve quickly and are willing to lock in supply and pricing years ahead.
What it means for buyers
For the rest of the technology economy, the shortage shows up in prices. Memory is a component in virtually every device and server, and its cost has been climbing as the shortage tightens. Smartphone and computer makers are passing those costs to consumers, and data-center operators are competing for allocation, sometimes receiving only a fraction of what they order. Mehrotra has said Micron’s data-center customers would take far more supply than the company can commit.
The longer the shortage runs, the more it reshapes where the industry’s profits land. Memory makers are capturing a share of the AI boom that was previously reserved for chip designers, and their margins have recovered sharply from the 2023 collapse. Wall Street has taken notice: analysts covering the sector have repeatedly raised price targets on memory names this year, and Micron’s shares have been among the strongest performers in the semiconductor group.
The supply response
The cure is capacity, and it is coming — slowly. Micron is investing heavily in new fabs, including large sites in Boise, Idaho and Syracuse, New York, and the industry’s combined capital spending has climbed sharply. But new memory fabs take years to bring online, and the next-generation products AI requires are more complex to manufacture than their predecessors. Mehrotra’s point this week was simple arithmetic: the demand curve is moving faster than the industry’s ability to bend its supply curve.
The shape of the demand is also different this time. AI systems do not just want more memory; they want the fastest memory, packed closest to the processor. High-bandwidth memory, the specialized chips stacked alongside AI accelerators, has become the industry’s hottest product line, and its supply is even more constrained than conventional DRAM. That is where the margins are, and it is why manufacturers are reallocating production toward data centers and away from the consumer electronics that once defined the industry.
Micron’s chief executive has put a date on the memory shortage: it does not end in 2026. The industry is betting its future on AI-specific products, locking customers into long-term contracts and rebuilding capacity that was never built during the last downturn. For buyers of servers, phones and everything else that uses memory, the takeaway is a pricing environment that stays tight for years. For memory makers, it is the best conditions the industry has seen in a generation — if the demand Mehrotra describes holds up.
