Samsung and SK Hynix are set to begin mass production of sixth-generation high-bandwidth memory (HBM4) chips next year, designed specifically for AI workloads. Samsung will start production in February 2026, while SK Hynix is expected to complete manufacturing by September 2026. Most chips will be used in Nvidia's Vera Rubin AI accelerator.
According to SEDaily, Micron and other competitors are unlikely to produce HBM4 in 2026, giving Samsung and SK Hynix a market advantage. Samsung's production will start earlier, in February. These chips are not aimed at the consumer market but are supplied to Nvidia. Samsung has passed Nvidia's quality tests, and SK Hynix is working with TSMC on a 12nm logic base die, while Samsung uses a 10nm process. HBM4 promises double the bandwidth and ~40% power efficiency gains, with better AI and non-AI integration.
Monthly production is expected to be 6,50,000 units for Samsung and 5,50,000 for SK Hynix, with HBM production at 1,70,000 and 1,60,000 units, respectively. All next-year HBM capacity is reportedly booked by AI companies, so consumer RAM shortages may continue through 2026.
Comments (0)
No comments yet
Be the first to comment!