Ahmed Shihab, chief product officer at Western Digital, argues that the real contest in AI storage is not about chasing the fastest medium. It is about whether capacity can keep expanding at a cost that remains affordable. His remarks were reported by ChainCatcher.
Shihab says many architectures run fine early on, but data growth from a few petabytes to hundreds of petabytes, or even exabytes, can expose problems through runaway cost. That cost, he argues, becomes an architectural issue at AI scale. Keeping bulk data on high-performance media for long stretches drains budgets for compute, networking, power and staff.
He draws a line between workloads. Flash fits model weights, GPU overflow, KV cache and session context — low-latency, high-performance jobs. HDD fits training corpora, logs, checkpoints, compliance records, synthetic data and inference output — bulky, long-lived and cost-sensitive data. His summary: “Flash handles the present; HDD handles the full lifecycle.”
For many batch workloads, the question is not whether flash can store the data, but whether customers can bear the cost of running everything on flash over time. Shihab does not see this as a flash-vs-HDD battle. Future AI storage, he says, will be tiered by performance, cost, power, density, reliability and data lifecycle. Matching the right medium to each workload from the start is what keeps the business sustainable as systems scale.
Ahmed Shihab, chief product officer at Western Digital, has laid out his view on what will actually decide AI storage competition, according to ChainCatcher. The core issue, he argues, is not raw speed. It is whether capacity can keep expanding at a cost that stays affordable as data grows.
Cost pressure arrives as data scales
Shihab says many architectures run normally in the early stage. The trouble shows up when data volume climbs from a few petabytes to hundreds of petabytes, or even exabyte scale. At that point, cost can spiral out of control instead of the technology failing at write or read level.
Storage cost itself becomes an architecture problem at AI scale. If bulk data sits on high-performance media for a long time, it eats into budgets that should go to compute, networking, power and people.
Flash and HDD have different roles
For low-latency, high-performance workloads — model weights, GPU overflow, KV cache and session context — Shihab says flash is the right fit. For large-scale, long-retention and cost-sensitive data, including training corpora, logs, checkpoints, compliance records, synthetic data and inference output, HDD is more appropriate.
His summary: “Flash handles the present; HDD handles the full lifecycle.”
For many bulk storage workloads, the bottleneck is not whether flash can store the data. It is whether customers can afford to keep using flash for all of it in the long run.
A layered approach, not a flash-HDD contest
Shihab does not frame the future as a direct competition between flash and HDD. AI storage, he argues, will be split into layers, decided by performance, cost, power, density, reliability and data lifecycle.
He emphasizes that the right medium needs to be matched to each workload from the start. Otherwise, as the architecture expands, business sustainability may be at risk.
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.
