Soaring AI-compute demand has tipped the global memory-semiconductor market into shortage, sending prices skyrocketing. A 16 GB + 512 GB module is expected to cost ~RMB 500 more by year-end. Phone makers are hardest hit: some can’t secure enough chips, production schedules are scrambled, and hot-model output is capped.
On Nov 6, TrendForce reported global memory chips in the steepest rally on record. DRAM: Upstream capacity is tight, and module houses such as Kingston are holding back shipments, triggering hoarding, which has lifted DDR5 chip spot quotes by 30% in a week and DDR4 chip spot quotes by 11.6%. NAND: supplier reluctance pushed 512 Gb TLC wafer spot up 14.2%. Supply strain will persist short term, prompting buyers to stock early.
SK hynix will be the sole source of 12-layer HBM3E for Google’s 7-gen TPU, while Samsung grabs the advanced-logic fab and the remaining HBM slots. 2026 TPU volumes are expected to double Samsung’s HBM shipments. Brokers tip SK hynix to hold 56.6 % of Google TPU HBM share in 2025 and Samsung 43.4 %. With Meta and AWS also ramping custom AI chips, server DRAM demand is seen +35 % YoY in 2026 vs. supply at only +20 %, keeping HBM and high-end DRAM prices on the rise.
Micron will stop all Crucial-branded consumer DRAM and SSD output by Feb 2026, pivoting 100 % to AI-datacenter HBM and enterprise products. Consumer margins are thin, and channels are messy, the company says, while demand for datacenter AI is surging. The exit further tightens retail supply and accelerates price inflation; Miccon’s freed capacity will be hurled into the HBM battle against SK Hynix and Samsung for AI-memory orders.
ASIC-server demand is booming. Foxcorn has locked in Google TPU v6p, AWS Trainium, and Microsoft Maia orders, becoming the top supplier to every major CSP with >40 % share. The company sees the segment maintaining hyper-growth and projects shipments to double in 2026, pushing the GPU-to-ASIC mix toward the market’s 8:2 split and creating a second growth curve after Nvidia GPUs.
Nvidia launched Alpamayo-R1, the world’s first vision-language inference model that gives cars human-like common-sense judgment. Vision specialist Solomon is first to adopt Jetson Thor + Isaac; its robot-inspection kit is already shipping to the Americas, and its AI sales are forecast to jump >70 % this year and next. MediaTek, meanwhile, is co-developing a 3 nm smart-cockpit chip (C-X1) with Nvidia, sampling H2, and set to add revenue in 2025.
