BlockBeats News, August 6th - Ahmed Shihab, Chief Product Officer of Western Digital, stated in a post that the key to AI storage competition is not simply pursuing the fastest medium, but whether the capacity can be continuously expanded at an affordable cost. He believes that many architectures work fine in the early stages, but when the data scale grows from single-digit PBs to hundreds of PBs or even EBs, potential cost issues may be exposed.

Flash memory is suitable for high-performance, low-latency scenarios such as model weights, GPU overflow, KV cache, and session context; HDD is more suitable for large-scale, long-term, cost-sensitive data such as training corpora, logs, checkpoints, compliance records, synthetic data, and inference outputs. He summarized it as, "Flash handles the present, HDD handles the entire lifecycle."

At AI scale, storage cost itself becomes an architectural issue. Storing large amounts of data long-term on high-performance media will encroach on budgets for computation, networking, power, and personnel. For many batch storage workloads, the issue is not whether flash memory can store, but whether customers can sustain the full cost of using flash memory long-term.

In the future, AI storage will not be dominated by a single technology, but should be designed in layers based on performance, cost, power consumption, density, reliability, and data lifecycle. He emphasized that this is not a competition between flash memory and HDD, but the need to match appropriate media for different workloads from the beginning; otherwise, the architecture may affect business sustainability as it scales up.