Sandisk went to the Future of Memory and Storage 2026 show in Santa Clara this month with a simple argument: AI’s next bottleneck is memory, not compute — and High Bandwidth Flash, not just HBM, is the fix.
The company’s FMS 26 presence (August 4–6, booth #607) covered next-generation NAND, High Bandwidth Flash and enterprise SSDs built for inference workloads: fast data lakes, model storage, retrieval-augmented generation and KV caches. It also updated its AI Data Cycle framework, which maps storage needs across the AI lifecycle.
Chief product officer Khurram Ismail argued that storing, moving and accessing data now matters as much as the computation itself.
“AI infrastructure has entered a new phase where the ability to store, move, and access data efficiently has become as important as compute itself,” said Ismail. “As inference workloads grow, customers need memory and storage technologies that can deliver ideal performance, capacity, and power efficiency without adding complexity.”
What Sandisk brought
- BiCS10 QLC NAND: next-generation 3D NAND aimed at density, performance and power efficiency for AI, cloud and datacentre storage. TheElec notes Sandisk shipped BiCS10 TLC samples in July.
- High Bandwidth Flash (HBF): a NAND-based memory technology engineered for AI inference at scale.
- Enterprise SSDs: high-endurance configurations for KV cache workloads, future PCIe Gen 6 support and ultra-high-capacity E3.S designs with reduced DRAM placement.
High Bandwidth Flash and the memory wall
Sandisk anchored the show with a keynote, “NAND: The Versatile & Scalable Foundation of the AI Era,” on Wednesday August 5, featuring Jim Elliott (chief revenue officer), Ismail and Alper Ilkbahar (chief technology officer). A day later it joined Google and SK hynix on the panel “Breaking the Memory Wall with High Bandwidth Flash.” The panel name is the thesis: NAND wants a seat at the memory table that HBM currently owns.
The groundwork is already public. On August 4, Sandisk and SK hynix published the first HBF specification through the Open Compute Project — covered here when it landed. SK hynix came to the same show with its tenth-generation 375-layer 4D NAND, claiming 2.5 times the power efficiency of its predecessor — the same week Samsung showed its own answer, putting AI processing inside the RAM itself.
For anyone watching memory prices, the direction matters: if HBF works, part of inference memory moves from DRAM and HBM onto cheaper NAND — token economics, in the keynote’s phrase. Flash is not about to replace HBM; it wants the parts of the memory hierarchy HBM is too expensive for. That is a bet Lenovo expects to keep paying for through the rest of the decade.
What is High Bandwidth Flash (HBF)?
High Bandwidth Flash is a NAND-based memory technology developed by Sandisk with SK hynix, engineered for AI inference at scale. The two companies published the first HBF specification in August 2026 through the Open Compute Project.
What is BiCS10 QLC NAND?
BiCS10 is Sandisk’s next-generation 3D NAND, built for higher density, performance and power efficiency in AI, cloud and datacentre storage. QLC stores four bits per cell; TheElec reports Sandisk shipped BiCS10 TLC samples in July 2026.
Will High Bandwidth Flash replace HBM?
No. Sandisk positions HBF as a complement to HBM: flash takes over the parts of the AI memory hierarchy that HBM is too expensive for, while HBM keeps the workloads that need its raw bandwidth.
Leave a Reply