Particle.news
Download on the App Store

Samsung and SK Hynix Set to Begin HBM4 Mass Production in February 2026 as AI Memory Squeeze Intensifies

Exploding AI demand is driving shortages with sharp HBM3E price increases.

Overview

  • Industry reports say Samsung will ramp HBM4 in February 2026, with most output slated for Nvidia’s next accelerator platform and some volume for Google’s next‑gen TPU.
  • SK Hynix is reported to deliver final 12‑layer HBM4 samples to Nvidia in early January, targeting formal volume production between February and March if validation completes on schedule.
  • Suppliers including Samsung and SK Hynix have raised 2026 HBM3E pricing by nearly 20%, an unusual move before a new generation takes over, reflecting sustained orders from AI chip buyers.
  • Procurement stress is escalating, with Korean media cited in Wccftech reporting a Google purchasing executive was fired after failing to secure long‑term HBM supply.
  • ASUS formally denied plans to build a DRAM wafer fab, saying it will rely on supplier partnerships and product adjustments rather than invest in memory manufacturing.