WD and multi-tiered disk drive arrays

A blog by WD’s Chief Product Officer Ahmed Shihab implied several roadmap possibilities which we explored with him via an email interview.

He made an initial statement: “Before getting into the specifics, I’d underscore that the core point of the blog that we believe is underrepresented in current coverage: AI infrastructure is shifting from a compute-centric model to a data-centric system.

“Much of today’s narrative focuses on GPUs and compute cycles. What’s less discussed, but increasingly critical, is how data is stored, managed, and retained at scale over time.

“That shift has real implications for architecture, economics, and long-term infrastructure design, and is the perspective we’re aiming to bring forward.”

Blocks and Files: You say “Data keeps moving behind the scenes to the lowest cost and is constantly being read and written for durability.” This implies object, tape or some other archival storage to me, which is not WD technology. In fact WD sold off its ActiveScale object storage business to Quantum several years ago. Is WD now considering providing disk or tape-based storage for applications needing high data durability?

Ahmed Shihab: The statement is about how object stores manage data durability which is to ensure the data read is always the same as the data written, even if that data is not touched by the end user. This is a fundamental process in all storage. At the cloud scale this is a significant problem, but also an opportunity to redistribute the data over more racks that may be a few more network hops away, to reduce the on-disk size of the erasure coded object at the expense of latency. This is a tricky balance as data access needs are not known before we need the data, especially as the latency can’t be longer than the SLA expected by the application.

This process means data is constantly moving around in object stores, so moving some data to archive is not an option since the latency gets into the seconds to hours range that violates the application SLA.

As data grows, we need HDDs that can keep up with this background scrubbing without affecting user traffic or wear out like Flash does. It is one of the reasons we are introducing high bandwidth drive technology.

At AI scale, this constant movement and durability management becomes a defining characteristic of the system, not a background function, which is why storage performance, bandwidth, and economics become increasingly critical.

Blocks and Files: You blog: "The systems that succeed will be those designed with this reality in mind: not as compute environments, but as data systems—where storage is foundational, architecture is tiered, and scale is defined by how effectively data is retained, managed, and used over time.” WD had a storage systems business in the past which it sold off. Eg, IntelliFlash. It has remnants with OpenFlex and RapidFlex but these are SSD-based. How do you envisage WD offering architecturally tiered storage products?

Ahmed Shihab: At its core, WD is a science, engineering, and nano-fabrication company. HDDs are complicated on the inside but easy to understand and use on the outside, so it’s up to us to innovate on behalf of customers as we have the tools and technology to do so. Our innovation is focused on understanding the customer needs in their tiered architecture and ensuring we build ahead of their needs to support the tiering and the economics they depend on. This is something that has been true for a long time, and we will continue to focus on our core strengths in HDD technology.

AI’s voracious demand for storage means we need to accelerate our innovation which is what we have done. We also realize that some of the innovation requires a lot of software to take advantage of that innovation, so we’re evolving our platforms to simplify adoption of those technologies - meaning we will support all customers who need to accelerate their journey to scale. In short, we are building the technology to support our customers’ storage and tiering needs with the best economics at scale.

The broader point is that AI infrastructure is no longer a single-tier problem. Systems are being designed across multiple storage tiers, and our role is to ensure those tiers scale economically and reliably as data grows.

Blocks and Files: Is WD going to develop/acquire technology to provide a tiering framework and/or provide colder storage than constantly spinning disk? For example, spun-down disk.

Ahmed Shihab: This is a question we get asked periodically. We actually have the technology to provide spun down disks today as we invest in technology ahead of need, but the operational needs of constantly moving data goes against spun down disks due to performance. With power optimized drive technology, we found a clever way to lower the power enough for it to be meaningful to customers without a significant loss in performance and a gain in capacity per disk. This formula is the first time we have seen interest in and positive feedback from customers in lower power technology as it is sympathetic to the software stack running above it.

What we are seeing is an emerging need for storage tiers that sit between high-performance and traditional archival, balancing access, cost, and power at scale. This is less about a single product and more about how architectures evolve to manage data across its lifecycle.

In practice, AI workloads require continuous data access and movement, which reinforces the need for always-available, power-efficient storage rather than traditional archival approaches.

Blocks and Files: In essence, what does the blog mean for WD's roadmap?

Ahmed Shihab: The blog highlights the role storage plays in the AI datacenter, and that at scale, it means HDDs (and SSDs) have to reinvent themselves as AI pivots to a data growth system rather than just a compute or model-focused effort. We are pleased with customer reaction to our roadmap, where some customers are engaging in faster adoption. More broadly, it reflects a shift in how we, and increasingly our customers, are thinking about AI infrastructure: not as a compute problem, but as a data system where scale, economics, and durability are the defining constraints.

In summary, we think this is an important perspective for the industry to engage with, particularly as it will directly shape how next-generation AI infrastructure is designed. Today’s AI conversation is largely centered on compute. The next phase will be defined by how effectively data is managed at scale.

Comment

The net of all this, as I see it, is that WD is making varieties of disk drives for tiered disk storage, for use by hyperscalers ad enterprises. All my initial ideas about Ahmed Shihab’s blog were mis-placed - which is good to know of itself. But there's not much of an impact on storage array-using enterprise customers here, until storage array suppliers and software-designed storage suppliers build systems using multi-tiered disk. The hyperscalers will build such systems themselves.