SK hynix (KOSE:A000660) and SanDisk Unveil Open HBF Specification for AI Inference Memory

SK hynix (KOSE:A000660) and SanDisk have published a new High Bandwidth Flash (HBF) technical specification through the Open Compute Project, an effort aimed at giving AI inference systems a shared, open framework for high-bandwidth memory and storage. The announcement is a clear sign that SK hynix wants to help set the agenda for AI infrastructure, not simply supply components into existing designs.

HBF is positioned as a complement to HBM, not a replacement. The proposed specification targets workloads that need both large memory capacity and strong data throughput, but do not always require the extreme speed of High Bandwidth Memory (HBM). With AI models continuing to grow in size, inference servers increasingly face a bottleneck in how quickly they can move data between storage, memory, and compute. HBF is aimed at that layer of the memory hierarchy, giving system designers more flexibility when balancing capacity, bandwidth, and cost.

The collaboration with SanDisk matters as well. SanDisk brings deep flash and storage expertise, while SK hynix brings DRAM and HBM experience. HBF sits between those worlds, and the involvement of both companies suggests the standard is intended to work across the memory hierarchy rather than in a single product niche. By pushing the spec through the Open Compute Project, the two companies are also signalling that they want broad industry adoption instead of proprietary lock-in.

For SK hynix, the timing is important. AI inference is becoming a larger share of data center workloads, and the company has already built a strong position in HBM, the advanced memory used inside AI accelerators. An open HBF standard could broaden that role, giving hyperscale and server vendors a more predictable way to integrate high-capacity memory and storage into future infrastructure. It also reduces the risk that SK hynix ends up tied to a closed or niche technology path as AI hardware evolves.

At the same time, open standards are not a guaranteed competitive moat. A standardized interface makes it easier for multiple players to build compatible products, which could expand the market but also invite more competition into SK hynix's core territory. The company will still need to differentiate through manufacturing quality, power efficiency, and close integration with customers.

What remains unclear is the cost of scaling HBF. SK hynix already has heavy capital commitments in South Korea, including its Yongin and Cheongju fab plans, and management has not yet disclosed how much additional investment HBF will require or how quickly customer adoption will develop. Investors are likely to look for more specific disclosure in upcoming conference presentations and quarterly earnings, including HBF capacity expectations, capex allocation, and how the technology fits alongside HBM in the product roadmap.

For now, the HBF announcement reinforces the broader investment story for SK hynix: AI demand for advanced memory is not limited to HBM, and the company is trying to shape how AI infrastructure is built, not just supply components into it. Whether that position translates into stronger financial performance will depend on execution, customer pull, and how capital-intensive the path to HBF actually becomes.