AI Memory Breakthrough
SK hynix and SanDisk unveil HBF, a new open memory standard to accelerate AI
Anonymous
Key Points
1
The HBF standardization consortium was launched in February, and the just-announced specification marks its first major milestone about six months after formation.
2
The collaboration that led to HBF began in August of last year and was formalized into an HBF standardization consortium in February, signaling a multi-stage, open development process.
3
The HBF specification was unveiled at the Future of Memory and Storage 2026 event, described as the world's largest memory and storage exhibition, held at the Santa Clara Convention Center in California.
4
Investor-focused coverage highlights that HBF is being pursued as an open standard via the Open Compute Project to encourage broad industry participation, with notes of an upcoming keynote and panel to discuss implementation and adoption.
SK hynix and SanDisk have released the first industry-standard specification for High Bandwidth Flash (HBF), a new memory technology built to sit between high-speed memory chips and traditional storage drives. The announcement was made at the Future of Memory and Storage 2026 event in Santa Clara, California, described as the world's largest memory and storage exhibition, according to Korea JoongAng Daily.
HBF stacks 8 or 16 NAND flash chips to deliver up to 512GB of capacity. It offers three bandwidth grades ranging from about 0.4 to 3.0 terabytes per second. The goal is to speed up AI workloads by closing the gap between fast but small high bandwidth memory (HBM) and large but slower solid-state drives, as Korea Herald reported.
HBF Fills the Gap Between Speed and Capacity
Modern AI systems are hitting a bottleneck. HBM chips are fast but hold very little data. SSDs hold a lot but move data slowly. HBF is designed to live in the middle. It uses stacked NAND flash — the same type of chip found in USB drives — but wires them together to move data much faster than a standard SSD, according to Korea Herald.
The specification uses the Universal Chiplet Interconnect Express (UCIe) interface. Think of UCIe as a universal plug that lets HBF chips connect to many different processors. That broad compatibility is key for getting HBF into AI servers from multiple vendors. The technology is also non-volatile, meaning data stays saved even when the power goes off — unlike HBM.
Open Standard Through Open Compute Project Draws Industry Partners
Rather than keeping HBF as a closed, proprietary design, SK hynix and SanDisk released the spec through the Open Compute Project (OCP), an open industry group. That choice invites other chipmakers, server builders, and AI companies to adopt the standard. Partners including Google and Tenstorrent are already involved, signaling early industry buy-in, Korea JoongAng Daily reported.
Yahoo Finance noted that the open standard approach is designed to encourage broad participation rather than lock customers into a single vendor. A keynote and panel discussion on HBF implementation and adoption are planned, giving the industry a formal forum to engage with the new spec.
Six Months From Consortium Launch to First Milestone
The collaboration between SK hynix and SanDisk began in August of last year. The two companies formally launched an HBF standardization consortium in February 2026. The spec released this week marks the consortium's first major milestone — roughly six months after its formation, according to Korea JoongAng Daily.
The fast timeline reflects urgency in the AI memory market. Data volumes in AI inference and agentic AI systems — where AI takes actions over time, not just answers single questions — are growing quickly. SK hynix and SanDisk are betting that getting there first with an open standard gives them an early leadership position.
What HBF Means for the AI Memory Race
HBM from SK hynix already powers chips like Nvidia's H100. But HBM is expensive and physically small. HBF could let AI systems keep far more data close to the processor without the high cost of HBM. At up to 512GB per stack and speeds reaching 3.0 TB/s at the top grade, HBF offers a compelling middle ground, Korea Herald reported.
Yahoo Finance highlighted that investor attention is growing around HBF as AI memory demand rises. With Google and Tenstorrent already at the table, the standard has momentum. The next step is getting server and chip designers to build products around the spec — a process that will play out over the coming months at events like OCP's global summit.
Sentiment
Mostly PositiveNEGATIVENEUTRALPOSITIVE
Publishers
13
Articles
25
Reach
38
Topics
A
AnonymousAuthor
