Sandisk Claims High Bandwidth Flash Can Rival HBM While Offering Much Larger Capacity
Sandisk is positioning its NAND-based High Bandwidth Flash, or HBF, as a potential breakthrough for AI and data-intensive computing. According to the company, the technology can deliver bandwidth comparable to High Bandwidth Memory while offering eight to 16 times more capacity at a similar cost.
That claim is significant because modern AI inference workloads are putting growing pressure on traditional memory systems. As artificial intelligence models become larger and more complex, processors need faster access to massive amounts of data. HBM has become a preferred solution for high-performance AI chips because of its speed, but it remains expensive and limited in capacity compared with flash storage.
Sandisk’s HBF aims to close that gap by bringing flash memory much closer to processors. Instead of treating NAND as slower storage sitting farther away from compute hardware, HBF is designed to provide high-throughput memory access that could better support demanding inference tasks. If Sandisk’s performance claims hold up in real-world deployments, the technology could offer a new balance of speed, capacity, and cost for AI infrastructure.
The biggest advantage appears to be density. Offering up to 16 times the capacity of HBM at a comparable price could make HBF especially attractive for workloads that require large model weights, frequent data access, and improved efficiency without dramatically increasing system cost. This could be useful in AI servers, edge computing hardware, accelerators, and other systems where memory capacity is becoming just as important as raw compute power.
HBM is still expected to remain critical for top-tier performance, especially in training massive AI models and powering high-end GPUs. However, Sandisk’s High Bandwidth Flash could create a new category of memory technology aimed at AI inference, where capacity, cost efficiency, and proximity to processors are increasingly important.
As the AI hardware market continues to evolve, memory innovation is becoming one of the key battlegrounds. Faster processors alone are not enough if they are held back by memory bottlenecks. Sandisk’s HBF proposal suggests that NAND flash may have a larger role to play in future AI systems, potentially bridging the gap between conventional storage and high-speed memory.
If successful, High Bandwidth Flash could help reshape how AI systems are built by giving developers and hardware makers access to much larger pools of fast memory without the steep cost and capacity limits typically associated with HBM.
