Memory Is The Lynchpin Of The IT Industry, And Doubly So For AI
AI-summarised brief · reviewed before publication
The IT industry has shifted from a compute‑centric focus—dominated by CPUs for five decades and GPUs for the last fifteen years—to a memory‑centric model driven by the rise of generative and agentic AI. Gartner’s latest chip revenue data shows memory (including DRAM, HBM and NAND flash) now outpacing non‑memory sales, with overall semiconductor revenue nearly doubling between 2025 and 2026 before a projected slowdown in 2027. Micron’s addressable market forecasts indicate that traditional DRAM categories will soon eclipse high‑bandwidth memory (HBM) sales. This surge is fueled by a massive upgrade cycle as aging server fleets are replaced, and by AI workloads demanding maximum core density, which pushes demand for high‑capacity DRAM and flash. Tight DRAM supplies, as manufacturers pivot to HBM for AI accelerators, have driven server‑grade DDR4 prices up 2‑3 × and DDR5 9‑13 × since 2022, while enterprise SSD costs have risen 6‑7 ×.
💡 Why It Matters
- · AI’s appetite for memory is forcing a wholesale redesign of data‑center economics, turning memory scarcity into the primary cost driver for modern computing infrastructure.
