Is the global memory shortage deliberate?
Mar 1st 2026
Memory prices have exploded in 2025 and 2026 as AI labs and hyperscalers ramp up purchases and chipmakers reallocate production, and disclosed wafer deals by OpenAI raise questions about whether some buyers are hoarding capacity to limit on-device AI and protect cloud-based business models.
- DRAM and NAND prices have spiked roughly 500 to 600 percent by February 2026, driven by sudden demand and tightening supply.
- OpenAI announced contracts with Samsung and SK Hynix for 900,000 memory wafers per month, a figure that industry sources and government officials have cited publicly.
- Our estimates show a fully built 10GW Stargate cluster would need about 3 million wafers, implying OpenAI’s disclosed purchases could exceed immediate needs by more than three times under common assumptions.
- Major makers paused secondary equipment sales and reallocated NAND capacity to DRAM, which compressed available supply and halted normal market refilling cycles.
- The shortage is hitting phones, PCs, gaming consoles, automotive systems, and GPUs with higher component costs, lower base memory in new models, and delayed product timelines.
- A plausible strategic effect of large wafer buys is to limit on-device inference capacity and preserve cloud inference revenue for hyperscalers and AI labs.
- Chipmakers and industry analysts forecast constrained supply through 2027 and into 2028 even as new fabs are planned or expanded.