Samsung Begins HBM4 Chip Shipments To Close AI Memory Gap With Competitors

Samsung Electronics has officially started shipping its next-generation high-bandwidth memory chips, HBM4 Chip, to customers as global demand from data centres powering advanced AI systems continues to surge. The South Korean tech giant announced the shipments on Thursday, though it did not disclose the identities of its initial clients.

HBM4 chips are designed to handle massive volumes of data at high speed, making them critical for modern AI processors. The new chips deliver a standard processing speed of 11.7 gigabits per second, a 22% improvement over the previous HBM3E generation, and can peak at 13 gigabits per second, easing data bottlenecks as workloads grow more complex.

Brand News Day Tech News Desk reports that the HBM4 chips, produced on a 4-nanometre logic process, offer capacities ranging from 24 to 36 gigabytes, with plans to scale to 48 gigabytes. The chips deliver up to 3 terabytes per second of bandwidth per stack, roughly 2.4 times that of HBM3E, alongside 40% better power efficiency and enhanced thermal management. These features are expected to support Nvidia’s upcoming Vera Rubin AI accelerator platform set to launch later this year.

Samsung also confirmed that samples of its next iteration, HBM4E, will be available in the second half of 2026. The move signals Samsung’s effort to catch up after trailing competitors in previous HBM generations, particularly in the high-speed memory market crucial for AI workloads.

The market for high-bandwidth memory is increasingly competitive. SK Hynix announced in January that its HBM4 chips are already in volume production, aiming to maintain a dominant market share. Micron has also moved early, reporting high-volume HBM4 production and initial shipments to customers.

Investor sentiment responded positively to Samsung’s announcement, with shares closing up 6.4% following the news. Analysts note that memory bandwidth has become as critical as processing power in modern data centres, reflecting a global push to expand AI infrastructure and maintain competitive advantage in next-generation computing.