Micron executives interpret: Why is it difficult to improve memory shortages before 2028 despite huge investments?

Micron executives stated that despite significant investments, the memory shortage issue will not improve before 2028. Christopher Moore, Vice President of Marketing for Micron Technology's Mobile and Client Business Unit, indicated in an interview that the current tight situation is not due to manufacturers actively adjusting their customer structures, but rather a combination of the explosive demand from AI data centers and the increasing complexity of memory manufacturing, which has rapidly tightened the industry's supply boundaries. In this context, even if industry capital expenditures remain high, substantial improvement in memory supply will still require a long time window, and the related tight situation may continue until around 2028. AI Consuming DRAM Capacity, Data Center TAM Rapidly Expanding The structural change in DRAM demand is key to understanding the shortage issue. The proportion of data center and AI-related demand in the entire DRAM market has rapidly risen from the past 30%-40% to 50% or even 60%. In this context, the entire industry is facing the reality of "insufficient capacity," rather than a resource allocation issue for a single manufacturer. Moore stated, this is not a challenge faced by Micron alone, but an industry-wide constraint that includes all major memory manufacturers. During the phase of concentrated expansion of AI data centers, no manufacturer can ignore this change in demand. He said: "This is not a Micron problem, but an industry problem. We and our peers or competitors are doing our utmost to serve these niche markets, but the supply is far from sufficient. This is indeed a regrettable situation." Consumers Have Not Been Abandoned, Just "Channel Forms Have Changed" In response to market doubts about Micron's "exit from the consumer market" after withdrawing the Crucial brand, Micron emphasized that its layout in the consumer end is more completed through OEM methods. The company still directly provides LPDDR and DDR modules to OEM manufacturers such as Dell and ASUS, and enters the end market through integrated systems. In Moore's view, the rise in priority for AI demand does not equate to consumers being marginalized, but rather a result of the total addressable market (TAM) expanding, while the supply side cannot keep up in the short term. Capacity Expansion Is Not About "Buying More Equipment," But System Constraints of Manufacturing Complexity Micron clearly denies the viewpoint that "simple capacity expansion can alleviate the shortage." The main bottleneck in current DRAM production is not the number of devices, but the production line switching losses caused by product specification fragmentation. As mobile and PC manufacturers increase their parallel demand for various capacities such as 8GB, 12GB, and 16GB, wafer fabs need to frequently switch between different designs (DID), which directly reduces effective output per unit time. After the explosion of AI demand, manufacturers actually need to reduce the variety of specifications and extend the continuous production cycle of a single product to maximize bit output. Moore stated: "You can imagine that if there are many different machines in a wafer fab producing a chip, and then you have to stop those machines to run on another chip, the yield will definitely drop. Of course, the actual situation is not that simple, but this is the best explanation I can think of. What we are doing now is to minimize the number of chips and minimize the number of different DID chips to maximize yield, understand? So we are working hard with our customers." The release pace of new capacity is slow, with a key turning point pointing to 2028 Moore revealed that the ID1 wafer fab that Micron started building three years ago in Idaho, USA, has advanced its production start time from the originally planned end of 2027 to mid-2027, but it will still take time for the new capacity to truly achieve large-scale shipments from equipment introduction, process ramp-up to customer certification. In his view, only when the new production line completes full certification and operates stably can there be a substantial change in the supply-demand relationship in the memory market, and this time point is more likely to fall in 2028. Moore further stated: Memory manufacturers are racing to build new production lines, but process limitations ultimately force them to advance their schedules by several quarters, which means that for ordinary consumers, the DRAM shortage may continue for quite a long time, or at least will not end until the demand for artificial intelligence begins to wane