AI's appetite for memory is reshaping the hardware market
The AI build-out is colliding with the memory supply chain. In a discussion with AI Magazine, SHI International's Cheryl Slater laid out how surging AI demand is driving DRAM shortages and rocketing RAM prices, stalling hardware deployment and inflating data-center costs - to the point that procurement leaders are having to rewrite their usual supply-and-demand playbooks.
What's driving the squeeze
- Memory makers are reallocating wafer capacity away from conventional DRAM toward high-bandwidth memory (HBM) for AI accelerators, and HBM consumes nearly three times the wafer capacity of DDR5 - worsening scarcity.
- SK hynix, Micron, and Samsung control most global DRAM production, and SK hynix reportedly secured demand for all of its 2026 RAM output back in October.
- Large AI data-center investment plans prompted hyperscalers to pull purchases forward and stockpile, draining the supply of server memory modules.
How long it lasts
The outlook is uncertain. Conservative estimates run into early 2027, but it hinges on several variables - how quickly fabs rebalance production, whether stockpiling is reined in, and whether some planned data centers stall over permitting or power-grid constraints, which could soften demand. Even if hyperscaler buying eases, lagging effects on pricing are likely as supply catches up.
What it means for buyers
The practical advice is to avoid panic buying - which risks locking in peak prices and worsening the shortage - and instead get strategic about memory efficiency and lifecycle planning. That includes prioritizing device refresh, optimizing server configurations to use memory more efficiently, and adopting more dynamic, continuous procurement rather than fixed refresh cycles. The broader takeaway is that the shock has undermined long-held assumptions - even the steady cadence of Moore's Law - and is pushing organizations toward genuine supply-chain resilience, with memory now reshaped from a commodity afterthought into a strategic planning concern.