SK hynix and Sandisk have initiated a joint effort to standardize High Bandwidth Flash (HBF), a next-generation memory solution designed to support the growing demands of AI inference.
On February 25 (local time), the two companies held a kickoff event at Sandisk’s headquarters in Milpitas, California, to announce their global strategy for HBF standardization. The initiative aims to establish HBF as an industry standard and strengthen the broader AI ecosystem.
To advance the effort, SK hynix and Sandisk will form a dedicated workstream under the Open Compute Project (OCP), a global open data center technology consortium. The workstream will focus on defining technical specifications and accelerating industry-wide collaboration.
The move comes as the AI industry shifts from model training to inference, where real-time service delivery requires fast and energy-efficient memory architectures. As more users access AI services simultaneously, conventional memory structures face limitations in handling large-scale data processing while maintaining power efficiency.
HBF is positioned between High Bandwidth Memory (HBM) and solid-state drives (SSDs), creating a new memory tier. It is designed to bridge the gap between HBM’s ultra-high bandwidth and SSDs’ large storage capacity, enabling both scalability and improved total cost of ownership (TCO) in AI inference workloads.
Industry observers expect demand for hybrid memory solutions incorporating HBF to grow significantly around 2030. As AI infrastructure competition increasingly depends on system-level optimization across CPUs, GPUs, memory and storage, companies capable of delivering integrated memory solutions are gaining strategic importance.
Leveraging their experience in HBM, NAND design and advanced packaging, SK hynix and Sandisk plan to accelerate both the standardization and commercialization of HBF.
Ahn Hyun, Chief Development Officer at SK hynix, said the future of AI infrastructure depends on optimizing the entire ecosystem rather than focusing solely on individual component performance. He added that HBF standardization would help build a collaborative framework and deliver optimized memory architectures for the AI era.
