SK hynix and Sandisk Release an HBF Specification, With Adoption Still to Prove

Sandisk describes the framework as a route for HBF to coexist with HBM, rather than replace it. In its release, the company says the aim is to give AI-inference-system and accelerator designers more near-compute capacity and bandwidth, improving power, performance and total cost of ownership. Those benefits are forward-looking company expectations, not measured results in the release.

Why an open standard matters—and what it does not establish

OCP is an open data-center technology initiative, and publication through its HBF technology workstream makes the specification available to other designers. That matters because memory technologies need matching processors, packages, system software and server designs before they can become a usable tier. An open interface may reduce one compatibility barrier compared with a proprietary attachment.

The two companies began their HBF standardization partnership in August 2025, formed the consortium in February 2026 and released this specification about six months later. Google and Tenstorrent joined as consortium members during that process; Sandisk says they contributed to technology validation and establishment of the standard. Their participation extends the work beyond the two primary contributors, but neither release identifies a commitment by those companies to deploy HBF in a product.

Sandisk says it wants HBF to become a de facto standard and to speed an early ecosystem. That makes the release a strategic bid to shape the design rules of a NAND-adjacent category, as well as a technical document. Its own forward-looking-statement section explicitly warns that expectations for standardization, adoption and the technology's benefits could differ materially from actual results.

Rendering of SK hynix’s exhibition booth at FMS 2026. Source: SK hynix Newsroom.

A separate NAND roadmap should not be mistaken for HBF proof

At the same FMS event, SK hynix said it would show a 375-layer, tenth-generation V10 4D NAND wafer and products under development. It says the generation improves performance per watt by 2.5 times over its predecessor and plans mass production of high-performance, high-capacity enterprise SSDs based on it early next year. Coverage of the event describes the same planned display and eSSD timetable.

That is a separate NAND-product roadmap. The reported performance-per-watt figure is SK hynix's comparison with its prior NAND generation; it is not an HBF benchmark, a comparison with HBM, or evidence that an HBF system will deliver the promised economics.

The evidence HBF needs next

The relevant next decision sits with accelerator, processor and server designers. They will need to build compatible systems and test whether a NAND layer adjacent to compute changes the cost-and-performance balance for particular inference workloads.

What would resolve the central claim is concrete evidence: shipping HBF devices and host systems; workload-level latency, throughput and endurance results; and like-for-like power and cost comparisons against the HBM-plus-SSD configurations HBF is intended to supplement. Until then, the OCP specification is a clear interoperability milestone, but its commercial value remains a proposition the hardware ecosystem must test.