Seagate 50TB HAMR drives: 2028, not 2027

Every headline this week put 50TB in the title, which is fair, because a ten platter drive holding fifty terabytes is a genuinely absurd object. Read the actual timing though and the drive lands in 2028, not next year: Seagate told investors on 28 July that the platform doing 5TB or more per disk starts qualification shipments in late calendar 2027, and qualification is the step before customers get real volume. So that's the correction. The part we'd actually plan around is buried in the same call, where the CFO said the gap between supply and demand is now a little bigger than a few quarters ago, and price per exabyte climbed about 10 percent.

Qualification is not shipping

Here is where the 50TB figure comes from. Seagate’s current platform, Mozaic 4+, tops out at 44TB and is in production with two of the largest cloud providers. The next generation moves past 5TB per disk. Stack ten of those platters in one enclosure and you land in 50TB territory, which is the arithmetic CFO Gianluca Romano walked analysts through.

Late calendar 2027 is when those drives go out for qualification.

That word does a lot of work. A hyperscaler takes a new drive, runs it for months against its own failure and firmware expectations, and only then signs a volume order. Nobody buys a nearline drive on a press release. So the honest reading is that 50TB is a 2028 product, and the 2027 date on the roadmap is the start of testing rather than the start of buying. TechSpot and Tom’s Hardware both landed on 2028 for shipments, and we think that’s right.

HAMR itself has stopped being the interesting question, by the way. It hit 40 percent of Seagate’s nearline exabyte shipments this quarter, and the company expects it to reach roughly 70 percent within a year. The heat-assisted recording that took a decade of skepticism is now simply how most of these drives get made.

The number nobody put in a headline

Price per exabyte rose about 10 percent year over year in the quarter. Seagate’s guidance for the September quarter implies closer to 20 percent.

Romano’s phrasing on the call was mild: the gap between supply and demand is now a little bigger than a few quarters ago. Read the margins and it sounds less mild. GAAP gross margin came in at 52.3 percent, on a product category that spent years grinding along in the thirties. Seagate is guiding the September quarter to $4.1 billion of revenue and non-GAAP EPS of $7.30, against $5.71 delivered in the quarter just closed.

Storage vendors do not post those numbers when there’s plenty of supply.

And the allocation picture matches. The vast majority of nearline exabytes are already spoken for into calendar 2028, with customers pushing to reserve 2029. Those are the big cloud contracts, not the drive you’d add to a NAS at home. But it’s the same media, the same heads, the same factories. When a manufacturer can sell every exabyte it makes to hyperscalers at a rising price, the retail channel is not where the discount shows up.

What we’d do with this

If you’re planning capacity for next year, plan it around 44TB and current pricing, not around a 50TB drive you cannot order. Roadmaps slip. Seagate’s own 30TB generation arrived later than first indicated, and this one has an extra qualification cycle in front of it.

Budget for storage getting more expensive per terabyte. That reverses about fifteen years of assumption, and it’s worth saying out loud to whoever signs your purchase orders before they see the quote.

Then there’s the design problem nobody enjoys. Drives keep growing and the interface does not. Seagate has published no sustained throughput number for the 50TB class, so run it as arithmetic: at 250 MB/s a single pass over 50TB takes roughly 55 hours. A parity rebuild reads every surviving member of the group as well, usually while the array is still serving traffic, usually slower than a clean sequential pass. Two of those drives failing in the same window stops being a thought experiment.

Which is why the shops we talk to keep moving to erasure coding with wider stripes, or to mirrored copies across failure domains, rather than trusting a classic parity rebuild to finish before something else breaks. Honestly, I think the capacity race has outrun the redundancy habits most teams still have.

One small thing worth checking before you size anything: a 50TB drive is 50 decimal terabytes, so your filesystem will report roughly 45.5 TiB. That’s a 9 percent gap, and on a rack of these it’s real money. We wrote up why MB and MiB disagree if you want the full version.

The AI connection, stated plainly

CEO Dave Mosley framed the demand as hyperscalers using hard drives to handle the long tail of AI workloads. That tracks. Training runs live on flash, but the training corpus, the checkpoints, the generated output nobody wants to delete, all of it lands on spinning rust because the cost per terabyte still is not close.

So the AI build-out shows up here as a hard drive shortage, the same way it shows up as a 50MW data centre moratorium in New York and as a memory price spike in phones. Different symptom, one cause.

Full year revenue was $12.2 billion, up 34 percent, with $3.1 billion of free cash flow. For a company that was widely written off as a melting ice cube a few years ago, that’s a strange place to end up.

Sources

Seagate’s own numbers come from the fiscal Q4 and fiscal year 2026 results published on 28 July 2026, on the Seagate investor relations site. The roadmap and pricing commentary comes from the earnings call the same day, summarised in this call breakdown. Capacity and timing reporting from TechSpot and Tom’s Hardware. The rebuild timing in this piece is our own arithmetic from an assumed sustained rate, not a Seagate specification.

Frequently asked questions