What Is High Bandwidth Flash (HBF) and Can It Overcome the AI Memory Wall?
HPCwire, Tuesday, August 4th, 2026
SK Hynix and Sandisk are developing vertically stacked flash to relieve the KV cache bottleneck in AI inference.
SK Hynix showcased high bandwidth flash, a new vertically stacked flash storage technology developed with Sandisk, at the FMS 2026 conference in San Jose.
Although still about a year from production, backers say HBF could help alleviate one of the biggest challenges in AI inference: the memory wall.
The key value cache has emerged as a major obstacle to rolling out AI services, since it stores pre-computed answers to previous queries and functions as short-term memory during inference.
Larger KV caches let AI service providers serve more and longer contexts, making cheaper high-bandwidth capacity a direct lever on inference economics.