News
Samsung and SK Hynix aim for high bandwidth flash commercialization by 2027 for AI GPUs
Sunday, January 18, 2026 at 08:31 AM
Experts predict High Bandwidth Flash (HBF) could be integrated into Nvidia, AMD, and Google GPUs by 2027 or 2028. Developed by Samsung and SK Hynix in collaboration with SanDisk, HBF aims to provide significantly higher capacity than HBM (up to 512GB) to address the memory demands of AI inference. SK Hynix is expected to demonstrate a trial version of HBF soon, targeting a bandwidth of over 1,638 GB/s.
Context
Samsung Electronics and SK Hynix are fast-tracking High Bandwidth Flash (HBF) for commercialization by 2027, targeting integration with Nvidia, AMD, and Google AI hardware. In collaboration with SanDisk, these firms aim to solve capacity bottlenecks inherent in DRAM-based HBM. SK Hynix plans to demo a prototype shortly, signaling a pivot toward NAND-based stacks that leverage existing HBM design expertise to accelerate time-to-market.
Technically, HBF delivers up to 512GB of capacity—nearly eight times that of HBM4—with bandwidth exceeding 1,638 GB/s. While its 100,000-cycle write limit necessitates software optimization for read-heavy inference, the architecture drastically streamlines data transmission by placing storage closer to the GPU. Industry experts suggest HBF adoption will surge alongside HBM6, potentially outgrowing the HBM market by 2038 as AI models demand increasingly massive datasets for real-time processing.
Related Companies
Nvidia
NVDA
SK Hynix
000660
Samsung Electronics
005930
Google
GOOGL
AMD
AMD