Navigating the New Memory Crisis: Insights from Lenovo at ISC 2026

Memory shortages, often referred to as the "RAMpocalypse," are reshaping how hardware buyers approach procurement in the tech industry. During the recent International Supercomputing Conference, Lenovo candidly remarked that memory prices "will never be like it was last year," emphasizing a shift towards a new normal that reflects ongoing demand pressures rather than a fleeting crisis.

Shifting Memory Economics

Lenovo's message draws attention to a significant transformation in the memory industry's economics. The decline in memory prices that characterized early 2025 is unlikely to return anytime soon, especially with consumer expectations and purchasing behaviors evolving around the growing demand for AI applications. Expectations surrounding increases in manufacturing capacities set to come online by 2028 won't likely alleviate pressure, as much of the added output is expected to be consumed by AI infrastructure. Companies are racing to ramp up their capabilities, and this arms race is changing not just prices but the fabric of the marketplace.

This context highlights a broader issue affecting the tech industry: the convergence of demand from various sectors is forcing a reevaluation of resource allocation. As more companies pivot towards AI, they're not just procuring memory, but also reshaping their procurement strategies to prioritize reliability and future-proofing over immediate cost savings. This shift is anticipated to have lasting implications, reshaping how businesses operate in a memory-constrained environment.

Production Capacity Insights

A noteworthy example supporting this trend comes from SK hynix, which announced a plan to triple its memory production capacity by 2034. This speaks to a long-term vision that diverges from expectations of returning to low-margin environments experienced in previous years. Memory manufacturers aren't likely to gamble on risky expansions if they see a looming return to oversupply and significant price drops. Instead, the focus appears to be on sustainable growth in response to persistent high demand, particularly from AI sectors.

This perspective highlights a shift not just in production goals, but in corporate strategies across memory manufacturers. The decision to expand production significantly reflects an understanding that immediate gains from inexpensive memory are outweighed by long-term stability and profitability. If you're in this space, you'll see that memory companies are banking on the AI wave driving demand for the foreseeable future.

Changing Server Architectures

Lenovo's presentation highlighted the increasing importance of memory capacity in server design and procurement. Historically, vendors focused on maximum supported memory capacities; however, populating server DIMM slots has now become prohibitively expensive, fundamentally altering procurement strategies. With today's dual-socket server architectures expected to launch next year, featuring 16 memory channels per processor, configurations of about 1 TB of installed memory are necessary to effectively harness their bandwidth.

This evolution in server architecture emphasizes that the tech landscape is changing rapidly. Now, if companies want to maintain a competitive edge, they must reevaluate their investments in memory. The need for massive memory capacities is not simply a trend; it's a necessity for organizations that plan to leverage high-performance computing effectively. As demand for speed and efficiency increases, the price of memory becomes just one piece of a much larger puzzle.

Broader Industry Impacts

Lenovo isn't navigating this landscape alone; other industry leaders share similar forecasts of enduring memory shortages. Micron forecasts that supply constraints will persist through at least 2027, with only marginal improvements expected thereafter. Likewise, SK hynix cautions that the memory shortage could extend to 2030, driven largely by the insatiable appetites of AI infrastructure. Micron's recent agreements worth roughly $100 billion with various customers underscore the seriousness with which hyperscalers approach long-term memory supply strategies.

Not even major players like Apple are immune from these supply challenges. Reports indicate that Apple has sought governmental approval to procure DRAM from CXMT, a Chinese manufacturer facing U.S. sanctions, highlighting the premium value placed on scarce memory resources. With memory vendors enjoying significantly higher pricing power and profit margins, there’s little incentive to revert to the volatile pricing cycles of the past. The memory market is becoming increasingly polarized, with implications that stretch across the tech sector.

Counterintuitive Trend: HBM's Rise

This ongoing memory shortage has paradoxically made High Bandwidth Memory (HBM) more attractive compared to traditional system memory. Through a shift in focus, DRAM manufacturers are reallocating production towards HBM for AI accelerator chips, which has unintentionally limited the supply of standard DDR5 and LPDDR5 memory, raising prices. Consequently, HBM, while still not cheap, now appears relatively more cost-effective for specific use cases due to the stark increases in conventional memory prices.

What's more, this equation isn’t just about dollars and cents. The choices organizations make now can have far-reaching implications for their future capabilities. Lenovo's insights suggest that GPU-accelerated computing models are becoming increasingly appealing for certain workloads, especially if applications can leverage GPU-attached HBM. By optimizing these configurations, organizations might find they can pare down on more expensive DRAM even while achieving notable performance gains. As traditional system memory now occupies a larger slice of overall server expenses, trimming memory capacity can translate into significant cost savings in infrastructure deployment.

Looking Ahead: The Future of Memory Procurement

While it's unclear whether Lenovo's longer-term projections will hold true, the cyclical nature of memory pricing cannot be ignored. The industry consistently experiences cycles of oversupply followed by dramatic downturns, and history may be poised to repeat itself. With hyperscalers investing heavily in AI infrastructure and memory manufacturers prioritizing high-margin enterprise products, the incredibly low prices of late 2024 and early 2025 might be an anomaly in retrospect. This creates an atmosphere of uncertainty for buyers.

And this is the part most people overlook: while companies focus on immediate resource allocation, the historical patterns suggest that today's pricing and supply restrictions could change in unexpected ways. What this means for you is clear: if you're working in this space, it's worth considering how you might adjust your procurement strategies in light of these shifts.

In the coming years, the memory market will likely remain a dynamic battleground. Companies that can effectively navigate these challenges stand to benefit immensely, while those caught unprepared may struggle to adapt. Keep your eyes peeled for developments; the implications of this supply-driven shift are just beginning to unfold.