HBF: more memory capacity for AI inference

What happened and what to do

A publication dated October 12, 2025, reports that SanDisk is proposing High Bandwidth Flash (HBF), a NAND-based design aimed at AI inference workloads. According to the text, the proposal targets 8 to 16 times HBM’s capacity, with similar read bandwidth and price; first samples are expected in the second half of 2026. To check the context and verify these details, consult the original publication by title and date and compare its specifications with any subsequent official information. These are expectations described in the publication, not results from a product already available.

For a company, this calls for infrastructure preparation, not immediate adoption of a technology whose samples are still expected. Teams can map memory requirements and bottlenecks in inference workloads, instrument usage and cost metrics, and create benchmark tests to assess future hardware options. Wendelmaques can help turn that assessment into purchasing, monitoring, and integration criteria without assuming HBF will suit every application.

How the consultancy can help

Wendelmaques can assess workloads’ memory, capacity, and performance requirements, define evaluation metrics and criteria, and propose a monitoring or benchmarking implementation with a client-specific delivery and operations scope.

Next step

Send a short description of your inference workloads and current memory limits. Wendelmaques can assess the case and provide a scoped proposal.