AI is still eating the extra wafers. According to Industry sources, DRAM contract prices could rise about 20% quarter on quarter in Q4 2026. NAND could rise about 30%. The seller’s market is intact. This is not a PC or phone refresh cycle. Makers cut capex and wafer starts in 2023-2024. That hole never fully closed. Then AI infrastructure soaked up what was left. NAND demand has shifted fastest. Phones and PCs used to set the mix. Now the pull is AI data centers, enterprise SSDs, and high-capacity QLC. Once large models move from training into inference, KV cache, RAG, and intermediate data all need fast, dense storage. NAND is becoming part of the AI memory stack, not only a consumer leftover. Makers are sending scarce bits to those higher-margin parts first. Weak handset and PC demand therefore hurts less than the old cycle would suggest. New supply is late. Kioxia’s K2, Solidigm’s second Dalian plant, and YMTC additions are coming online. Large greenfield fabs do not add much until after 2028. On DRAM, HBM4, server DDR5, high-capacity RDIMM, SOCAMM, and LPDDR keep taking advanced wafers, which squeezes commodity DRAM. Samsung’s P4 and SK Hynix’s M15X are the nearer expansions, and both lean toward HBM and advanced DRAM. $MU $SKHY Micron’s Idaho fab, SK Hynix Yongin Y1, Micron’s Taiwan Tongluo site, and Nanya’s 5A plant mostly contribute after 2027. Taiwan names in the slipstream are Nanya and Winbond on DRAM and specialty memory, plus Phison, ADATA, Transcend, and Team Group on controllers and modules, via higher prices, denser SKUs, and inventory marks. Price is doing the work because bit growth cannot. That lasts until the 2027-2028 fabs actually ship.