AI’s Memory Crisis: A Decade of Soaring Storage Prices

A looming memory and storage crisis threatens to reshape the technology landscape for years, driven by the extraordinary demands of artificial intelligence. Industry analysts and manufacturers are issuing stark warnings about severe shortages of NAND flash and DRAM, predicting a decade of escalating prices for components like SSDs and memory modules. The primary catalyst is the voracious consumption of these resources by AI data centers, which are now claiming a dominant share of global production capacity.

For PC builders, the last couple of years offered a welcome period of affordability. Solid-state drive prices plunged to unprecedented lows in 2023, making high-performance NVMe models nearly as inexpensive as basic mechanical hard drives. DRAM followed a similar path, reaching price points not witnessed in almost ten years. However, 2024 marked a definitive reversal, with costs for both NAND and DRAM beginning a steady climb.

This market shift stems partly from the cyclical nature of memory production, but the current situation is intensified by the unique needs of AI and large-scale cloud providers. A widespread supply squeeze now affects every segment of the industry. Whether you’re looking at consumer-grade SSDs, DDR4 memory kits, enterprise storage systems, or bulk hard drive shipments, the trend is unmistakable: prices are rising across the board in a synchronized manner not seen in the market for a long time.

The transition from market surplus to scarcity has been rapid. The downturn in 2022 and early 2023 placed memory producers in a difficult financial position, with NAND and DRAM selling below manufacturing cost and inventories swelling. Companies reacted with significant production cuts to stop the financial losses. By late 2023, the effects of these reductions reached the sales channels. Spot prices for 512Gb TLC NAND components, which had previously hit record lows, more than doubled over a six-month period, with contract pricing moving in the same direction.

Consumers felt this rebound almost immediately. By early 2024, a 2TB Western Digital Black SN850X was selling for over $150, while the price for a 2TB Samsung 990 Pro jumped from a holiday discount of about $120 to more than $175 in a similar timeframe.

The DRAM market’s recovery trailed NAND by about three months, but the trajectory was identical. DDR4 modules, which seemed like clearance items in 2023, encountered a supply crunch as manufacturers began phasing out their production. Projections indicated that prices for PC-grade DDR4 products could surge by 38-43% quarter-over-quarter by Q3 2025, with server-grade DDR4 expected to rise 28-33%. Even the graphics memory market felt the strain. As vendors moved to GDDR7 for new GPUs, shortages of GDDR6 pushed its prices up by approximately 30%. DDR5, while still ramping up as the new mainstream standard, also showed a clear and steady price increase.

Hard drives were not immune to these pressures. Western Digital informed its partners in April 2024 of plans to raise HDD prices by 5-10% due to constrained supply. At the same time, industry analyst firm TrendForce highlighted a shortage of nearline HDDs, the high-capacity drives used extensively in data centers. This shortage has forced some data center workloads to shift onto flash storage, thereby tightening the NAND supply even further.

The unprecedented demands of artificial intelligence are placing immense strain on global memory supplies. Every major memory cycle has a catalyst; in the past, it was the proliferation of smartphones, then solid-state notebooks, followed by cloud storage expansion. The current cycle is overwhelmingly fueled by AI. Training and operating large language models consumes colossal amounts of memory and storage. A single GPU node within a training cluster can utilize hundreds of gigabytes of DRAM and several terabytes of flash storage. The cumulative figures within large data centers are staggering.

Recent developments underscore the scale of this demand. OpenAI’s “Stargate” project is reportedly in discussions with Samsung and SK hynix for a monthly allocation of up to 900,000 DRAM wafers. That volume alone would represent nearly 40% of the entire world’s DRAM output. Whether this specific deal materializes fully is less important than the signal it sends about how aggressively AI companies are securing supply on a massive scale.

Cloud service providers are adopting a similar strategy. High-density NAND products are effectively sold out for months in advance. Samsung’s next-generation V9 NAND is almost fully booked even before its official launch. Micron has reportedly presold the vast majority of its High Bandwidth Memory production through the year 2026. Procurement contracts that once covered a single quarter now extend for multiple years, with hyperscalers purchasing directly from the source.

The ripple effects are reaching everyday consumers. Raspberry Pi, which had built up memory reserves during the market downturn, was compelled to increase its prices in late 2025 due to rising memory costs. The company’s CEO, Eben Upton, stated officially that “memory costs roughly 120% more than it did a year ago.” It appears no segment of the market can avoid the pricing surge.

Manufacturers are fundamentally shifting their investment priorities, which exacerbates the supply crunch. A shortage isn’t solely about demand skyrocketing; it’s also about supply being deliberately redirected. Over the last ten years, NAND and DRAM producers learned that rapid, unchecked production expansion typically leads to a market collapse. The oversupply that followed each boom period devastated profit margins, so the industry’s response in this cycle has been far more cautious.

Major players like Samsung, SK hynix, and Micron are all channeling capital expenditure toward High Bandwidth Memory and advanced manufacturing nodes. HBM offers exceptionally high profit margins, making it a clear strategic focus. With Micron’s entire 2026 HBM output already committed, every wafer allocated to HBM is one less wafer available for standard DRAM. A parallel shift is happening in NAND production, where engineering and manufacturing efforts are concentrated on 3D QLC NAND designed for enterprise clients.

According to the CEO of Phison Electronics, a leading NAND controller company, this redirection of capital spending will lead to tight supply conditions for what he believes could be the next ten years. He cited two primary reasons: the historical pattern where heavy investment led to price collapses and unrecovered costs, and the significant reallocation of investment by major manufacturers into HBM due to its attractive margins, leaving less capital for flash memory development.

These strategic decisions are putting a tighter squeeze on mainstream products. Production of DDR4 is being scaled back faster than demand is decreasing. Similarly, TLC NAND, once widely available, is now being rationed as manufacturers focus their resources on the most profitable segments, leaving older but still critical product lines undersupplied.

The storage market is experiencing a unique convergence of constraints. For the first time, both NAND flash and hard disk drives are in short supply simultaneously. Historically, when one became expensive, the other provided a cost-effective alternative. However, training large AI models involves processing petabytes of data, all of which requires a storage home. This “warm” data typically resides on nearline HDDs in data centers, but demand is now so intense that lead times for the highest-capacity drives have extended beyond one year.

With nearline HDDs difficult to obtain, some hyperscalers are fast-tracking the deployment of QLC flash arrays. While this addresses one bottleneck, it creates another by transferring demand pressure back onto the NAND supply chain. We are now seeing SSDs being adopted at scale for applications where cost-per-gigabyte previously made them impractical. The outcome is a supply squeeze from both directions, with HDD prices climbing due to limited availability and SSD prices firming up as cloud buyers enter the market to secure their needs.

Building new manufacturing facilities is not a simple or quick solution. New fabrication plants, or fabs, are under construction, but they represent a colossal investment and require a long timeline to become operational, particularly in the United States. A new, ground-up memory fab carries a price tag in the tens of billions of dollars and needs several years to reach volume production. Even expanding existing production lines involves months of installing and calibrating sophisticated equipment, with key suppliers like ASML and Applied Materials facing significant order backlogs.

Manufacturers are also cautious about repeating past errors. If demand cools or if companies pause procurement after building up stockpiles, an overbuilt market could trigger another price collapse. The memory of the painful downturns in 2019 and 2022 remains vivid, making companies hesitant to make large bets on long-term cycles, even as AI demand appears limitless today, especially with many observers believing the current AI boom may be a bubble.

Geopolitical factors add another layer of complexity to expansion plans. Export controls on advanced chipmaking equipment and restrictions on rare earth elements create hurdles for any potential HDD factory expansion. Hard drives depend heavily on Neodymium magnets, which are among the most in-demand rare earth materials. The HDD industry is one of the world’s largest consumers of these magnets, and China, which currently dominates their production, has recently restricted supply as part of ongoing trade disputes.

Even with sufficient capital, the supply chains for the necessary manufacturing tools and raw materials are themselves constrained. A shortage of skilled semiconductor engineers further slows progress. The overall result is a market characterized by deliberate discipline, where manufacturers prefer to sell their existing supply at higher profit margins rather than risk instigating another market collapse.

This cautious approach from manufacturers is unlikely to change in the near future. For consumers, it means the era of ultra-cheap PC upgrades is over. Enterprise clients, meanwhile, must prepare for larger infrastructure budgets. Storage systems, servers, and GPU clusters all require more memory at a higher cost. Many large cloud providers manufacture their own SSDs using custom controllers, while other major companies, like Pure Storage, purchase NAND in enormous volumes for all-flash arrays that power AI data centers. Some hyperscalers have adapted by securing their supply years in advance. Smaller operators without that kind of leverage face longer wait times and significantly higher costs.

Flexibility is diminishing across the board. Consumers might postpone an upgrade or settle for lower capacities, but the broader impact is a slowdown in the adoption of high-capacity drives and larger memory configurations. Enterprises have little option but to absorb the increased costs, given the fundamental importance of memory for AI and cloud computing tasks.

The market will eventually find a new equilibrium, though predicting when is difficult. New fabs are being built with government support, and if demand growth slows or procurement pauses, the cycle could swing back toward oversupply. Until that rebalancing occurs, prices for NAND flash, DRAM, and hard drives are expected to stay high through 2026. Enterprise buyers will continue to receive priority allocation, leaving consumers to compete for the remaining supply. The seasonal price drops that were once a reliable feature of the market are not likely to reappear anytime soon.

(Source: Tom’s Hardware)