A week ago, a blog was published called “The Flash Emperor Has No Clothes,” arguing that the all-flash array industry built a decade-long pitch on a premise that was never true and has now publicly collapsed. The piece was deliberately provocative.
This week, two earnings reports made the case.
The Receipts
On April 28, Seagate (STX) reported Q3 FY26 revenue of $3.1 billion, up 44% year over year. Non-GAAP gross margin hit 47%, an all-time high. Data center demand accounted for 80% of revenue, and nearline drives, the high-capacity HDDs sold predominantly into hyperscale and cloud customers represented close to 90% of total exabyte shipments. Seagate raised annual growth guidance to at least 20%. The vast majority of nearline capacity is already allocated through fiscal 2027.
The same week, Western Digital (WDC) reported Q3 FY26 revenue of $3.3 billion, up 45% year over year. Cloud revenue grew 48% YoY and represented 89% of WDC’s total revenue. Gross margin exceeded 50%. Multi-year customer agreements now extend into 2028 and 2029.
Combined, these two HDD vendors did $6.4 billion in revenue last quarter. Roughly 85% of that (over $5.4 billion) came from hyperscalers, cloud service providers, and large data center customers. The contracts behind those numbers run for years.
This is what the hyperscaler vote looks like in market-data form. The largest data infrastructure operators on earth are not just buying HDDs they are buying them at record rates, on multi-year commitments, while continuing to invest aggressively in NVMe flash for performance tiers.
If “flash is the new HDD” had ever been true, you would not see this. You would see HDD revenue declining, not growing 44–45% year over year. You would not see $6.4B quarters and contracts running through the end of the decade.
The Hyperscalers Run Mixed Fleet. Period.
This is the architectural pattern visible in two public earnings calls in a single week. The leaders run mixed-fleet, software-defined storage: high-performance NVMe flash for hot data and metadata, high-density HDD for capacity, all unified by software they either built or licensed. Just enough flash to saturate workload performance, then HDD for everything else.
This is not a hedge. It is not a transition state. It is the steady-state architecture for the largest AI and cloud workloads on the planet, and the multi-year contract durations confirm it. The hyperscalers are committing through 2028 and 2029, past at least three more potential AI-driven NAND cycles, to a mixed-fleet design.
The Detail Nobody Is Talking About: Which HDDs They’re Buying
The HDDs Western Digital and Seagate are shipping in record volumes are not the kind of drives that go into legacy hybrid architectures. They are not dual-ported SAS drives. They are:
- Single-port SATA – the commodity interface used in software-defined, shared-nothing architectures. No HA controller dependency, no proprietary backplane, no firmware lock-in.
- HAMR (Heat-Assisted Magnetic Recording) – Seagate’s Mozaic 4 platform. Highest capacity drives in the market. Seagate’s earnings call explicitly stated HAMR exabyte output is expected to become dominant by late 2026.
- SMR (Shingled Magnetic Recording) – including Western Digital’s UltraSMR family. WDC’s earnings call called out SMR adoption specifically as a margin and TCO driver.
These three categories — single-port SATA, HAMR, SMR — share one important architectural property: they are designed for software-defined, shared-nothing storage systems. They are not the drives legacy hybrid architectures were built around. Those systems require dual-ported SAS for controller failover. They cannot natively consume, at scale, the drives the hyperscalers are buying by the exabyte.
In other words: the hyperscalers are buying record volumes of exactly the drive classes that legacy hybrid architectures cannot natively consume.
This is the second-order proof point of the mixed-fleet thesis. It is not just that the hyperscalers are buying HDD. It is that they are buying the kind of HDD that confirms a specific architectural model: software-defined, shared-nothing, designed around commodity media. By buying at hundreds-of-exabytes scale, they are pulling the entire industry’s HDD roadmap in that direction. The next decade of HDD innovation will be built for the architecture VDURA HYDRA shares with the hyperscalers, not the legacy hybrid architectures built in the 2010s.
Two Weeks, Two Confirmations
Ten days ago, an open letter from one all-flash CEO admitted the central economic premise of all-flash had run out of road for the foreseeable future. This week, $6.4 billion of public earnings revenue — concentrated almost entirely in hyperscaler and cloud customers, with contract durations extending into 2028 and 2029 – showed who is already operating on the next chapter, and what they are buying to build it.
Mixed fleet is the present. The contracts running through 2029 say it is the future as well. The architectural question for any organization buying AI or HPC storage in 2026 is no longer whether to embrace mixed fleet, but how quickly to deploy a platform that does it natively, on the same drive classes the leaders are buying.
The architecture argument is over. Mixed fleet won the market, $6.4 billion of revenue this week, hyperscaler contracts running through 2029, and a roadmap of single-port SATA, HAMR, and SMR drives that legacy hybrid architectures cannot even consume. The only question left is how quickly the rest of the industry catches up.
