Memory chip makers signal multiyear supercycle with margin expansion through 2027

Memory chip makers, Micron, SanDisk, SK Hynix, and others, have begun signaling what they are calling a supercycle for NAND and DRAM driven by generative AI and agentic AI workloads that require vastly more silicon than traditional computing. Early-stage AI data centers had a CPU-to-GPU ratio of roughly 1:8, but as agentic systems mature, that ratio is shifting closer to 1:1, meaning exponential CPU and memory demand. Gross margins for memory suppliers are now exceeding 70 percent, and manufacturers are guiding for sustained high pricing through 2027 as supply tightness persists. Micron alone posted 190 percent revenue growth guidanceCompany-issued forecasts of future financial performance. against JPMorgan's 5, 10 percent guidance, a disparity traders see as evidence of structural AI tailwinds. The narrative is so dominant that even contrarian skeptics concede the near-term demand reality while questioning sustainability. Some point to the circular capex trap: Nvidia is funding infrastructure providers like CoreWeave to solve GPU scarcity, but if those capex booms eventually flood supply, margin compression could arrive faster than expected. This concern is reflected in the record 2.4 billion dollar inflow into the 3x Semiconductor Bear ETFExchange-Traded Fund - a basket of securities trading like a single stock. (SOXS) in April, the largest monthly intake ever, suggesting sophisticated traders are hedging tail risks of a sharp correction. Conversely, bullish advocates argue the AI infrastructure buildout has years to run and that memory makers will see price stickiness because compute density demands are inelastic. Earnings reports from Micron and SanDisk over the next month will be critical prove-it moments for supply-chain durability and demand elasticity.