On August 4, 2026, at the Future of Memory and Storage (FMS) conference in Santa Clara, SK hynix and Sandisk published the first official HBF technical specification through the Open Compute Project (OCP) . Key highlights:
The spec defines three performance grades :
| Grade | Read Bandwidth |
|---|---|
| Entry | ~0.4 TB/s |
| Mid | ~1.6 TB/s |
| High | ~3.0 TB/s |
Sandisk has stated its first-generation HBF will target read bandwidth up to 1.6 TB/s, roughly equivalent to one HBM4 stack but at approximately 10x the capacity .
The HBF consortium is led by SK hynix and Sandisk. Two major industry players have publicly joined:
Additional participants have not been officially named, but Samsung is reported to be developing its own HBF technology (see competitive landscape).
The HBF space is emerging, with three distinct approaches:
| Company / Approach | Technology | Capacity / BW Targets | Timeline | Notes |
|---|---|---|---|---|
| SK hynix + Sandisk (HBF consortium) | Open OCP standard; TSV-based NAND stacks | 512 GB / up to 3.0 TB/s (Gen 1: 1.6 TB/s) | Samples H2 2026, devices early 2027 | Open spec; Google & Tenstorrent onboard |
| Samsung | Proprietary high-bandwidth NAND (unnamed internally) | Targeting similar capacity and bandwidth to HBF | Targeting late 2027 – early 2028 for Nvidia/AMD/Google products | Developing in parallel; not part of the SK hynix/Sandisk consortium |
| Kioxia | "GP Series" XL-Flash based; PCIe 5.0 connected SSD modules | 5 TB module / 64 GB/s bandwidth ; roadmap to 100M IOPS | GP Series SSDs from Q3 2026; 100M IOPS by 2027 | Different architecture — still a PCIe-attached SSD, not a memory-bus device; developed at Nvidia's request |
| Nvidia | Not developing HBF directly, but acting as the key qualification gate and customer | N/A | Working with Kioxia on ultra-fast SSDs; evaluating Samsung's HBF for future GPUs | Nvidia is the critical certification gate for any HBF product to reach the GPU ecosystem |