SK Hynix and SanDisk reveal first specs for High Bandwidth Flash — but don't get too excited, you probably won't be able to buy it anytime soon
SK Hynix and SanDisk published the first High Bandwidth Flash (HBF) technical specification via the Open Compute Project. It targets up to 512GB stacks with three bandwidth grades around 0.4TB/s to 3.0TB/s using UCIe, with consortium partners including Google and Tenstorrent. Samples are expected in 2H 2026 and AI inference devices in early 2027, with commercialization later.
How this was made
The 30-second read
Why it matters
The disclosure is a concrete technical milestone with defined capacity and bandwidth grades, plus a stated sampling timeline for memory and inference devices. However, the article repeatedly emphasizes that HBF is not yet commercially available and that commercialization likely lands in late 2027 to 2028, with broader data center justification closer to 2030.
Market read
Traders get a new, specific technical milestone and roadmap windows for HBF, but the lack of near-term availability and the long commercialization runway limit immediate trading impact.
What to watch
The article notes Nvidia is not in the named consortium; partner testing, interface qualification, and production ramp execution could be the real bottlenecks rather than the published specs.
Background
SK Hynix and SanDisk published initial High Bandwidth Flash (HBF) specifications through the Open Compute Project, positioning it as a UCIe-connected, NAND-based high-bandwidth memory alternative for AI inference systems.
Ticker impact
SK Hynix published the first High Bandwidth Flash (HBF) specification, but provided no HBF commercialization date in the document.
Limited immediate impact; any stock reaction is likely tied to broader memory/AI infrastructure sentiment rather than a near-term catalyst.
The article discloses a new open specification and roadmap elements, yet explicitly says HBF exists only on paper for now and gives no mass-production timing for HBF itself.
SanDisk (Western Digital) released the first HBF technical specification via OCP, with samples of HBF memory expected in 2H 2026 and inference devices in early 2027.
Low near-term price impact; medium-term optionality may support sentiment among AI memory/infrastructure investors.
The article includes specific sampling windows and a commercialization range, but also notes the technology is not yet purchasable and depends on hyperscaler adoption and partner testing.
Market effects
Could shift parts of the AI memory stack narrative from HBM-only toward higher-capacity NAND-based alternatives, but the article stresses HBF is not a near-term HBM substitute.
Primarily affects global AI hardware supply chains; no specific regional demand shock is described.
If adopted by hyperscalers, HBF could influence global data center memory interface roadmaps and procurement planning over the 2027 to 2030 window.
Counterpoint
HBF may remain a niche standard because it is capacity-focused and still requires hyperscaler validation, while HBM supply and pricing dynamics could keep dominating near-term AI accelerator designs.
Key entities
- companySK Hynix
Co-published the first HBF specification; provided no HBF commercialization date in the document.
- companySanDisk
Co-published the first HBF specification; roadmap includes HBF memory samples in 2H 2026 and inference-device samples in early 2027.
- organizationOpen Compute Project
Platform through which the HBF specification was published openly.
- technology_standardUCIe
Open chiplet interconnect standard used to connect HBF stacks to host processors.
- companyGoogle
Joined the HBF consortium, supporting partner validation prospects.
