As the artificial intelligence (AI) industry reorganizes around ultra-large models, interest in next-generation data processing technologies is surging. Amid this trend, SanDisk's 'HBF (High Bandwidth Flash)' is gaining attention as a new AI memory solution.

Currently, AI servers operate with GPUs and HBM as their core axes. However, as model scales expand rapidly, demand for high-capacity memory capable of storing and processing more data is increasing quickly. The industry is accelerating the development of a new tier of memory technology to complement existing architectures.

HBF aims to simultaneously secure large-capacity storage capabilities and high data processing efficiency by stacking multiple NAND flash chips at high density. The concept is to reduce bottlenecks in AI workloads by utilizing NAND—which had previously remained a simple storage device—in a position closer to the computing unit. In particular, its core strength is cited as the ability to more efficiently manage large-scale data and model parameters during AI training and inference processes.

The industry expects that HBF will offer higher performance than existing SSDs while remaining competitive in terms of capacity expansion. To build the related technology ecosystem, SanDisk is currently discussing standardization centered around the Open Compute Project (OCP) together with SK hynix. This is interpreted as a strategy to flexibly develop technical specifications in line with the rapidly changing AI infrastructure environment.

Recently, in AI data centers, not only processor performance but also data movement speed and storage efficiency have emerged as key factors determining overall system competitiveness. The next-generation memory market is also seeing a shift in the center of gravity from computing power-centric to data supply capability-centric.

The industry is focusing on the possibility that HBF will serve as a complement that expands the memory hierarchy of AI infrastructure, rather than a replacement for HBM. There are also predictions that future AI systems will evolve in a direction where HBM and HBF are utilized together, each taking on different roles.

[This article was written with AI assistance. This article is not a recommendation to invest in any specific company or technology, and responsibility for investment decisions and outcomes lies with the investor.]