HDDs Are Having Their Best Year in a Decade: Most of the Enterprise Channel Hasn't Noticed

The HDD Renaissance

While the storage industry was busy panicking about flash, the HDD industry quietly rewrote its own roadmap.

In the last four months alone:

• Western Digital announced it is qualifying 40 TB UltraSMR ePMR drives at two hyperscalers, with volume production in H2 2026. The same roadmap extends to 60 TB on ePMR and 100 TB on HAMR by 2030.
  
• WD unveiled High Bandwidth Drive (HBD) technology — simultaneous multi-head read/write that delivers 2× drive bandwidth today, with a roadmap to 4×, 6×, and 8×. Translation: a nearline HDD that ships at 500+ MB/s now, headed for 1,000–2,000 MB/s in a few years. That is SATA-saturating, flash-class bandwidth on a $15–20/TB medium.
  
• WD also showed dual-pivot actuators — a second independent actuator on a separate pivot inside a 3.5-inch enclosure. Another 2× bandwidth. Drives in the lab today, GA in 2028.
  
• Toshiba began sampling its M12 Series, 30–34 TB host-managed SMR with FC-MAMR and its first glass-substrate platters. 282 MB/s data rate, 18% lower watts-per-TB, 11 platters in a 3.5-inch helium enclosure. CMR up to 28 TB sampling Q3 2026. HAMR and 12-disk drives coming in upcoming quarters.
  
• Seagate continues to lead capacity-per-platter (Mozaic / HAMR + SMR at 36 TB GA today, 50 TB by 2026, 120 TB by 2030 on the published roadmap).

In aggregate, this is more HDD innovation in 24 months than the industry delivered in the previous decade. And almost nobody outside the hyperscalers is set up to use any of it.

The “SMR Paradox”

Here is the uncomfortable truth: virtually 100% of the new high-capacity drives are SMR — host-managed, sequential-write, requiring software that speaks the drive’s native physics. Hyperscalers have figured it out. Dropbox runs ~90% of its storage on SMR. Google and Wasabi built their own zone-aware stacks.

The enterprise channel? Less than 1% SMR adoption today.

It’s not a hardware gap. The drives are here. WD is shipping. Toshiba is sampling. Seagate is GA. The gap is the software layer between those drives and the customer. Legacy filesystems and SDS stacks were architected for 1950s random-access logic, and they treat a modern 32 TB SMR drive exactly the same way they treated a 4 TB CMR drive a decade ago — by throwing small random I/O at a medium that physically wants long sequential streams.

The cost of that mismatch is enormous. Stacks running on HDD hardware without proper zone awareness waste 30–60% of the potential economics of the hardware — capacity, performance, and drive lifespan combined. They generate excessive seek operations, write amplification, and head wear that ages a drive prematurely. They blow tail latency budgets and turn AI training runs into intermittent torture.

This is the actual reason your 32 TB drives feel “slow.” It is not the drives.

The performance reset nobody is pricing in

Combine two things that just shipped:

On the hardware side, WD’s HBD turns a 26 TB nearline drive that previously delivered ~250 MB/s into one that delivers 500+ MB/s today and is on a public roadmap to 1,000–2,000 MB/s. SATA link saturation is now a real engineering concern for HDDs, which is a sentence I never expected to write.

On the software side, Leil OS 5.x — paired with a high-performance SAS4 24G HBA like the ATTO ExpressSAS H240F — achieved 10.5 GB/s sequential writes (EC 6+2) and 11.4 GB/s sequential reads across 60 Host-Managed SMR drives in a single server chassis. These aren’t theoretical projections; they are fully validated, published benchmark results.

In other words: the IOPS-per-TB conversation that defined storage architecture for two decades is being reset, on both sides of the stack, simultaneously. The HDD vendors are delivering bandwidth their drives never had, and the software layer (when designed for it) is unlocking 99.7% of theoretical maximum throughput.

Why the channel hasn’t moved (and what it costs them)

I’ll be blunt about the structural problem. SMR drives sit on a shelf because most enterprise storage stacks lack the support to make them usable. Customers who can’t find the solution to use them, or use them as not part of the close source product — and that’s a large fraction of the public sector, sovereign clouds, and OSS-first enterprises — simply walk away from the drives entirely. Meanwhile, hyperscaler-grade efficiency stays a hyperscaler privilege.

The cost of this gap shows up in three places:

• Capacity tax. Without zone-aware software, you leave 20–25% of usable capacity on the table on the same hardware. That’s a +25% “free” capacity gain Leil’s customers see going from a generic SDS to HDD-native.
  
• Performance tax. Without sequential-optimized data flow and Command Duration Limits, you fight tail latency for every minute of an AI training run. Drives hang on difficult sectors for tens of seconds; legacy software just waits. (Our answer: SNIA-standard CDL with a 50 ms hard deadline and parity-reconstruction fast-fail. Drive misses the deadline, we reconstruct from parity instead of waiting).
  
• Operational tax. Adding petabytes shouldn’t require six months of tuning by PhD-level staff. The hyperscalers automated this. The channel inherited the manual labor.

What “vendor-agnostic, drive-type-agnostic” actually means

Here’s the thing that I think gets understated when storage software is discussed: most software-defined storage products are tied to a single vendor’s drives, a single drive type, or a single firmware generation. That’s a tax on your future. The HDD industry is moving fast. Drive types are diverging — ePMR and HAMR will share a chassis at WD. FC-MAMR and HAMR will coexist at Toshiba. Seagate is mixing recording technologies across its product lines. SMR is going from optional to default.

Any architecture you commit to in 2026 has to absorb that diversity without forklift upgrades.

Leil’s stack — Leil FS (open core) and Leil OS (enterprise distribution) — was built to be exactly that:

When the WD-Toshiba-Seagate roadmap gets noisier — and it will — your software shouldn’t care.

An open invitation

Two things to close on.

First — to my fellow storage software vendors: we publish our test setups, our erasure-coding schemes, our durability models, and our performance numbers. We invite you to do the same. Better yet, let’s run a public benchmark. One rule: the results go public. No NDAs. No selective disclosure. If your stack does what it says on the brochure, this should be easy.

Second — to the infrastructure buyers reading this: the drives are ready. The roadmap is real. The software gap is closeable. You do not need to be a hyperscaler to operate like one. You need a stack that was designed for the physics, not retrofitted to it.

The next decade of storage belongs to the people who actually read the drive datasheet.

Author: Alexander Ragel
Alexander Ragel is the CEO of Leil Storage, an Estonia-based storage infrastructure company building HDD-native software for the exabyte era. Leil is in active co-engineering partnership with Western Digital, validated across the WD, Seagate, and Toshiba drive ecosystems.

Sources

1. Western Digital Innovation Day 2026 announcements (Feb 2026)
   
2. Toshiba M12 Series press release (Apr 2026)
   
3. Blocks & Files coverage of WD HBD and dual-pivot technology
   
4. Seagate Mozaic roadmap (publicly disclosed)
   
5. SaunaFS 5.0 release notes
   
6. Leil Press Briefing (April 2026)