--- title: "Why Are Memory Manufacturers Not Rushing to Expand Production?" type: "News" locale: "en" url: "https://longbridge.com/en/news/285376660.md" description: "A Bernstein report points out that NAND prices have surged more than 20-fold, far exceeding the economic threshold for replacing HDDs, thereby driving some demand back to mechanical hard drives. Despite high prices, NAND manufacturers are cautiously expanding production due to prolonged losses and insufficient returns, while HDD oligopolies are actively maintaining supply discipline. The tight balance between supply and demand is difficult to resolve in the short term, suggesting that strong memory prices may persist, even as AI reshapes storage architectures and technological pathways" datetime: "2026-05-06T12:23:19.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/285376660.md) - [en](https://longbridge.com/en/news/285376660.md) - [zh-HK](https://longbridge.com/zh-HK/news/285376660.md) --- # Why Are Memory Manufacturers Not Rushing to Expand Production? AI-driven storage demand has surged, with NAND prices skyrocketing more than 20-fold. Yet, both flash memory manufacturers and hard drive makers remain restrained in their large-scale production expansion. Behind this lies a supply discipline constructed from economic logic, technical barriers, and industry memory. According to Zhuifeng Trading Desk, a recent industry expert webinar hosted by Bernstein Research highlighted how the surge in NAND prices is reshaping storage architecture choices in data centers. Robert Soderbery, former Executive Vice President of Western Digital and head of SanDisk’s flash business, pointed out that **the current price gap per GB between NAND and HDDs has widened to approximately 20 to 25 times, far exceeding the 2-to-3-times threshold required for hyperscale cloud providers to switch based on total cost of ownership (TCO). This makes replacing HDDs with NAND economically unattractive.** This widening price gap is prompting AI data center operators to reevaluate their storage architectures, with some demand potentially shifting back from NAND to HDDs. Meanwhile, **NAND manufacturers, having endured years of loss cycles, are generally cautious about expanding production, while HDD oligopolies are actively maintaining supply discipline.** For investors, this implies that supply constraints for both types of storage assets will be difficult to break in the short term, and the pattern of strong prices may continue. ## AI Reshapes Storage Architecture; NAND Demand Concentrated on Latest Process Nodes Data center storage architecture has evolved from the traditional "two-tier model" to an AI-driven "three-tier architecture." Robert Soderbery introduced that traditional data centers used small-capacity enterprise SSDs for computing tasks and nearline HDDs for capacity storage, with HDDs accounting for about 80% to 85% of total bit volume. The rise of AI data centers has disrupted this landscape. The data preprocessing stage in AI workloads—including pre-computation tasks such as vectorization and embedding—places extremely high demands on storage performance, which traditional nearline HDDs cannot meet. This has spawned explosive demand for high-capacity enterprise SSDs. Capacity specifications have jumped from the previous 2–8 TB to 32 TB, 64 TB, and even 256 TB, with NAND accounting for up to 60% to 70% of new AI storage deployments. However, NAND demand is not evenly distributed across all production capacity. Robert Soderbery pointed out that the high-performance, high-capacity NAND required for AI workloads must use the latest process nodes (currently generations 7 to 9). Due to continuous pressure on NAND manufacturers and insufficient capital expenditure in recent years, the capacity of these latest nodes accounts for only about 30% to 35% of the industry's total capacity. The high concentration of demand on this limited capacity is the core reason for the sharp rise in NAND prices in this cycle. ## Insufficient Returns on Expansion Lead NAND Manufacturers to Wait Why are NAND manufacturers not accelerating production expansion in the face of soaring prices? The answer lies in the fact that return on investment remains unattractive. Robert Soderbery explained that the NAND industry has experienced continuous losses over the past five to nine years, leaving manufacturers with deep scars from the cycle and little confidence in the sustainability of demand. It typically takes about 15 months from capital expenditure to capacity realization—about a year for equipment delivery and several more months for debugging and commissioning. This means that prices must continue to rise and remain elevated for several quarters before manufacturers gain enough confidence to initiate expansion decisions, after which it will take another five quarters for new capacity to be released. More critically, even before this round of price increases, NAND manufacturers' gross margins were only around 20%, a level of return entirely insufficient to support large-scale capital investment. Robert Soderbery pointed out that the current focus of NAND manufacturers' expansion is on migrating existing capacity from older generation nodes to more advanced ones to better serve the high-performance demands of the AI market, rather than simply increasing total capacity. The industry typically targets an annual bit growth rate of 20% to 30%, with a significant portion of this growth coming from density improvements brought by QLC technology, rather than an increase in actual wafer input. ## HDD Oligopolies Actively Maintain Supply Discipline Unlike NAND, supply constraints in the HDD market stem more from active choices than technical bottlenecks. Currently, the HDD market effectively has only two major suppliers: Western Digital (WDC) and Seagate Tech, with Toshiba having relatively limited competitiveness. Robert Soderbery stated bluntly, "The first rule of oligopoly is not to expand production until target profit margins are achieved." HDD manufacturers have ample motivation to maintain this discipline. Over the past decade, the HDD industry has long been in a state of overcapacity, making manufacturers highly wary of repeating past mistakes. Furthermore, the inherent complexity of HDD technology—covering HAMR (Heat-Assisted Magnetic Recording) technology, read/write heads, mechanical structures, and materials science—creates considerable friction costs for expansion, objectively supporting supply constraints. Robert Soderbery revealed that Western Digital has more PhD researchers on the HDD business side than on the NAND business side, highlighting its technology-intensive nature. This means HDD manufacturers should be viewed as oligopolies capable of maintaining reasonable returns over the long term, rather than purely cyclical stocks. ## Long-Term Threat of NAND Replacing HDD Is Overestimated The market has long worried that NAND will gradually erode HDD market share, but this logic faces significant obstacles at the economic level. Robert Soderbery calculated that for NAND to substantially replace HDDs at the TCO level, the price gap would need to narrow to within 2 to 3 times. To reduce NAND costs to this level, the industry would need to complete iterations across multiple process nodes, with each node upgrade requiring about $50 billion in capital expenditure. Achieving effective replacement of HDDs might require cumulative investments of around $150 billion—and this does not include the capital needed to serve the AI market. Meanwhile, the cost reduction curve for NAND is slowing down. NAND has shifted from two-dimensional planar expansion to three-dimensional stacking, where the marginal cost benefits of increasing layers are diminishing, slowing the pace of cost reductions. In contrast, HDD HAMR technology is accelerating cost reductions, causing the cost curves of the two to converge. Robert Soderbery believes that the timeline for NAND to pose a substantial substitution threat to HDDs is "still quite distant." ## Transformation of Long-Term Agreements Improves Industry Planning Visibility Another important change in the storage industry is that long-term purchase agreements (LTAs) are shifting from historical "unilateral contracts" to truly bidirectional constraints. Robert Soderbery introduced that past LTAs essentially only constrained suppliers—suppliers had to reserve capacity, but customers could refuse pickup if not needed, while suppliers faced liability for breach of contract if they failed to deliver. This asymmetric structure made it difficult for manufacturers to conduct effective capital planning. SanDisk recently disclosed that it had signed five new types of long-term agreements, involving approximately $42 billion in purchase obligations, of which $11 billion was financial guarantees. Robert Soderbery evaluated this positively, believing that such bidirectional constraint contracts can significantly boost manufacturers' investment confidence and help the industry emerge from the predicament of insufficient supply. However, he also noted that if the market experiences a significant downturn, customers still have the incentive to renegotiate by paying penalty fees; the actual binding force of the contracts depends on market conditions. Using SanDisk's data as an example, the $11 billion guarantee accounts for about 25% of the total purchase obligations, meaning that even if prices fall by 50%, the effective price decline would be narrowed by about 25%. ## HBF: A New Frontier for NAND in AI Computing Power Beyond the evolution of storage architecture, High Bandwidth Flash (HBF) technology represents a potential opportunity for NAND to penetrate the AI computing layer. HBF draws on the packaging technology of High Bandwidth Memory (HBM), adopting a stacked structure and wide bus interface aimed at co-deployment with GPUs. Its core advantage lies in leveraging the specific characteristics of AI inference workloads—during model inference, large amounts of coefficients are streamed into the GPU rather than being frequently read and written. Although NAND has slower dynamic read/write speeds and higher latency, once streaming begins, the throughput is considerable, fitting this scenario perfectly. By combining HBM-like bandwidth with NAND's higher storage density and differentiated cost structure, HBF possesses potential competitiveness in AI inference scenarios. Robert Soderbery considers HBF "very promising," but notes it is still in the early stages. Whether it can meet commercial requirements in terms of power consumption, performance, and reliability remains to be verified. He stated that regardless of whether HBF can directly replace HBM, the space for innovation by NAND in the AI pipeline remains vast, and investors can view it as a potential upside option. ### Related Stocks - [MU.US](https://longbridge.com/en/quote/MU.US.md) - [STX.US](https://longbridge.com/en/quote/STX.US.md) - [AB.US](https://longbridge.com/en/quote/AB.US.md) - [WDC.US](https://longbridge.com/en/quote/WDC.US.md) - [6502.JP](https://longbridge.com/en/quote/6502.JP.md) - [WDCVV.US](https://longbridge.com/en/quote/WDCVV.US.md) - [6588.JP](https://longbridge.com/en/quote/6588.JP.md) ## Related News & Research - [Seagate Technology Holdings (STX) Net Margin Strength Tests Bearish HDD Narratives In Q3 2026](https://longbridge.com/en/news/284739470.md) - [Seagate and Western Digital Are a Hard Disk Drive Duopoly. Barchart Ranks the Storage Stocks Here.](https://longbridge.com/en/news/285102343.md) - [Amazon launches B2B supply chain service](https://longbridge.com/en/news/285245286.md) - [Micron vs. SanDisk: Which stock is the better buy for the AI boom?](https://longbridge.com/en/news/284976473.md) - [Seagate Stocks’s Rally Is No Accident: 3 Reasons Wall Street Is Optimistic](https://longbridge.com/en/news/285217264.md)