Pull to refresh
Logo
First open standard published for high bandwidth flash AI memory

First open standard published for high bandwidth flash AI memory

New Capabilities

SK hynix and Sandisk release an Open Compute Project spec for a flash-based memory tier meant to ease AI memory bottlenecks, with Google and Tenstorrent backing it.

Today: First open HBF standard released at FMS 2026

Overview

AI chips are starving for memory. On August 4, 2026, SK hynix and Sandisk published the first open technical standard for a possible fix: High Bandwidth Flash, or HBF.

The design stacks cheap flash memory (NAND) next to the processor. It promises up to 8 to 16 times the capacity of today's premium AI memory at a similar price. Google and the chip firm Tenstorrent have joined the effort, a sign the idea has real industry backing.

Why it matters

If HBF works, AI chips could hold entire large models right next to the processor for far less money, cutting the cost of running AI.

Questions about this story

No questions yet — be the first to ask.

Key Indicators

512 GB
Capacity per HBF stack
The spec defines stacks holding up to 512 gigabytes, using 8-high and 16-high NAND die stacks.
3.0 TB/s
Peak data rate
Top bandwidth defined in the standard, close to what high-bandwidth memory delivers today.
8–16x
Capacity vs. DRAM-based HBM
How much more data an HBF stack can hold than a same-size stack of high-bandwidth memory.
4
Companies backing the standard
Sandisk and SK hynix lead; Google and Tenstorrent joined the effort.

Voices

Curated perspectives — historical figures and your fellow readers.

Ever wondered what historical figures would say about today's headlines?

Sign up to generate historical perspectives on this story.

Play

Exploring all sides of a story is often best achieved with Play.

Most of these play right now — no account needed. Sign up to save scores, keep a streak, and unlock Debate and Predict. Log in Sign Up
Predict 4 ways this could play out. Back the one you believe — contrarian picks score more when a scenario has a resolution date. Log in to play

People Involved

Organizations Involved

Timeline

August 2025 August 2026

3 events Latest: Today
  1. First open HBF standard released at FMS 2026

    Today Milestone

    SK hynix and Sandisk publish the first HBF technical specification through the Open Compute Project. Google and Tenstorrent join the effort. It defines up to 512 GB per stack and about 3.0 TB/s.

  2. HBF work formally starts inside the Open Compute Project

    Standardization

    The two companies begin global standardization under a dedicated HBF workstream at the Open Compute Project.

  3. Sandisk and SK hynix agree to standardize HBF

    Partnership

    Sandisk brings in SK hynix to co-develop High Bandwidth Flash as a shared industry standard, not a single-vendor product.

Historical Context

3 moments from history that rhyme with this story — and how they unfolded.

2013–2015

High-bandwidth memory standardized (2013–2015)

SK hynix and AMD co-developed high-bandwidth memory (HBM), stacking DRAM chips next to a processor for far faster data flow. JEDEC published the standard in 2013. The first product shipped in AMD's Fiji graphics chip in 2015.

Then

HBM was expensive and niche at first, used mainly in high-end graphics and specialized computing.

Now

HBM became the default memory for AI accelerators, and SK hynix's early lead turned into a dominant, high-margin business.

Why this matters now

HBF follows the same playbook: a memory maker and partners set an open standard early, hoping to own the next AI-memory tier.

July 2015 – 2022

Intel Optane and 3D XPoint (2015–2022)

Intel and Micron launched 3D XPoint, sold as Optane, a memory tier meant to sit between fast DRAM and slower flash storage. It promised near-DRAM speed with far more capacity.

Then

Optane found some use in data centers but never reached the volume needed to bring prices down.

Now

Intel wound down the Optane business in 2022 after years of losses, and the between-tier product line was discontinued.

Why this matters now

HBF also fills the gap between fast memory and bulk storage. Optane shows that a clever middle tier can still fail if the economics do not work.

March 2019

Compute Express Link consortium formed (2019)

Intel and a group of chip and cloud firms launched Compute Express Link (CXL), an open standard for connecting processors to pooled memory. It grew to include most major hardware vendors.

Then

Early CXL products were limited, and adoption took several years to build.

Now

CXL became a widely backed way to expand and share memory across servers, absorbing rival efforts along the way.

Why this matters now

Like CXL, HBF is being pushed as an open, multi-vendor standard. Broad backing early is meant to avoid a format war and speed adoption.

Sources

(8)