The tech industry’s long-running memory shortage shows no signs of easing, and Lenovo is urging businesses to accept the new normal. Speaking at the International Supercomputing Conference 2026, Lenovo representatives delivered a sobering assessment: memory prices will not return to the unusually low levels seen in early 2025. While the comment carried a touch of humor—execs reportedly smirked as they made the claim—the underlying message was unmistakable. The era of cheap, abundant RAM and storage is over, and companies must adapt their strategies accordingly.
A permanent shift in memory economics
Lenovo’s presentation, titled The 5 Step RAMaggeddon Survival Guide, underscored a fundamental change in the memory industry’s dynamics. The company argued that even as new manufacturing capacity comes online starting around 2028, demand from AI infrastructure will likely absorb much of the additional output. This imbalance between supply and demand suggests that DRAM and NAND prices may stabilize at higher levels than in previous cycles.
The reasoning is simple: memory manufacturers like SK hynix, which plans to triple its production capacity by 2034, are unlikely to invest billions in expansion if they expect a return to the razor-thin margins and oversupply conditions of early 2025. Historically, the memory market has been cyclical, swinging between shortages and gluts. But the rise of AI workloads—particularly those requiring massive memory bandwidth—has disrupted that pattern.
Memory capacity becomes a critical cost factor
Beyond pricing, Lenovo highlighted how memory capacity itself is reshaping server design and purchasing decisions. Vendors have long emphasized the maximum supported memory in new platforms, but actually populating those slots has become prohibitively expensive. A key trend is the introduction of dual-socket servers with 16 memory channels per processor, which will arrive in 2027. Even modest configurations could require up to 1 TB of installed DDR5 memory to fully leverage the available bandwidth.
This shift has significant implications for IT budgets. Enterprises can no longer treat memory as a secondary consideration; it has become a primary cost driver in server deployments. Lenovo’s guidance suggests that businesses must prioritize memory efficiency in their infrastructure planning, balancing performance needs with budget constraints.
Industry-wide forecasts paint a grim picture
Lenovo is not alone in warning of a prolonged memory crunch. Micron has told investors that supply constraints will persist at least through 2027, with incremental improvements expected beginning in 2028. SK hynix has gone further, suggesting shortages could extend into 2030 as AI infrastructure continues to consume wafer capacity.
These forecasts are backed by multi-year supply agreements worth approximately $100 billion that Micron has already secured with major customers. Even industry giants like Apple are feeling the squeeze, reportedly lobbying the U.S. government for access to memory chips from Chinese manufacturer CXMT—a Pentagon-blacklisted firm—to alleviate supply pressures. The move underscores the desperation to secure additional DRAM amid rising prices.
Memory vendors are now enjoying some of the strongest pricing power and profit margins in years. This financial incentive reduces their urgency to return to the boom-and-bust cycles that once defined the DRAM market. The result is a structural shift where supply remains tight, and prices stay elevated.
HBM emerges as a strategic alternative
An unexpected consequence of the memory shortage is the growing appeal of High Bandwidth Memory (HBM) over conventional system DRAM. Manufacturers have redirected significant production capacity toward HBM for AI accelerators, reducing the supply of commodity DDR5 and LPDDR5 despite persistent demand.
This shift has narrowed the price gap between HBM and traditional system memory—not because HBM has become cheaper, but because DDR5 has become dramatically more expensive. For hyperscalers, this creates an incentive to maximize GPU utilization by keeping working data in HBM, thereby reducing the need for large DDR5 installations in host systems. Lenovo suggests that GPU-accelerated computing may now offer better cost efficiency for certain workloads, as applications can leverage HBM’s superior bandwidth while minimizing reliance on system DRAM.
What’s next for the tech industry?
The message from Lenovo and other industry leaders is clear: the memory shortage is not a temporary blip but a long-term reality. Businesses must reassess their infrastructure strategies, prioritize memory efficiency, and explore alternative architectures like HBM-accelerated computing. While supply conditions may improve slightly after 2028, the era of cheap, abundant memory is over. Companies that fail to adapt risk facing escalating costs and operational bottlenecks in the years ahead.
AI summary
Lenovo, RAM fiyatlarının yakın zamanda düşmeyeceğini ve AI talebinin bellek arzını sürekli olarak zorlayacağını açıkladı. Piyasa tahminleri ve üretici stratejileriyle ilgili detaylar burada.
