Western Digital Chief Product Officer Ahmed Shihab said the real contest in AI storage is not about choosing the fastest medium available, but about whether capacity can keep scaling at a cost customers can continue to bear. He argued that some storage architectures work well at the start, yet begin to break down economically when data volumes move from a few petabytes to hundreds of petabytes or even exabyte scale. In his view, storage cost becomes an architectural issue at AI scale.
Shihab said flash is better suited to high-performance, low-latency workloads such as model weights, GPU spillover, KV cache and session context. HDD, by contrast, fits large, long-retention and cost-sensitive data sets including training corpora, logs, checkpoints, compliance records, synthetic data and inference outputs. He summarized the division this way: flash handles the present, while HDD handles the full lifecycle.
He added that keeping massive data sets on high-performance media for the long term can consume budget that would otherwise go to compute, networking, power and staff. Rather than a contest between flash and HDD, he described future AI storage as a layered design problem shaped by performance, cost, power use, density, reliability and data lifecycle.
Western Digital Chief Product Officer Ahmed Shihab said AI storage should not be built around the fastest medium alone, arguing that the central question is whether capacity can keep expanding at an affordable cost.
In a post published on Aug. 6, Shihab said many architectures work in their early stages, but their weaknesses can surface once data scales from a few petabytes to hundreds of petabytes or even exabytes because costs can spiral. At AI scale, he said, storage cost itself becomes an architecture issue.
Flash and HDD serve different workloads
According to Shihab, flash is well suited to high-performance, low-latency workloads including model weights, GPU spillover, KV cache and session context. HDD is a better fit for large-scale, long-retention and cost-sensitive data such as training corpora, logs, checkpoints, compliance records, synthetic data and inference outputs.
He summed up the distinction in one line: “Flash handles the now, HDD handles the whole lifecycle.”
All-flash designs can strain budgets over time
Shihab said storing large volumes of data on high-performance media over long periods can crowd out spending on compute, networking, power and personnel. For many bulk storage workloads, the issue is not whether flash can store the data, but whether customers can afford to keep everything on flash over time.
He added that future AI storage will not be dominated by a single technology. Instead, it should be designed in layers based on performance, cost, power consumption, density, reliability and data lifecycle. In his view, this is not a fight between flash and HDD. The task is to match the right medium to the right workload from the beginning, or the architecture may start to affect business sustainability as scale increases.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
