SK hynix said on June 17 that it has shipped samples of HBM4E, its next-generation high-bandwidth memory for AI, to major customers, pulling the timeline forward from a previously guided second-half schedule.
What the 12-layer HBM4E delivers
The 12-layer samples reach 48GB per stack, run up to 16Gbps per pin and deliver roughly 4TB/s of bandwidth, with more than 20% better power efficiency than HBM4. SK hynix says HBM4E improves on the prior generation through changes to the DRAM core die, interface design and manufacturing process, key levers as AI accelerators demand ever more memory bandwidth.
Race with Samsung heats up
The early sampling tightens SK hynix's competition with Samsung, which sent its own samples on May 29. SK hynix led the HBM market with a 58% share in the first quarter of 2026, according to Counterpoint Research, and the company is moving into customer qualification as demand from NVIDIA and other accelerator makers surges.
Memory at the center of AI economics
HBM has become a strategic chokepoint for AI compute, a dynamic underscored by NVIDIA and SK hynix's multiyear memory pact and the broader scramble for advanced packaging and capacity seen across deals like Qualcomm's pursuit of Tenstorrent.
Reporting based on coverage from PR Newswire and SK hynix.