HAMR Hard Drives: What Is Shipping Now and What Is Coming

What changed and why

Heat-assisted magnetic recording briefly warms the spot being written with a laser, which allows a smaller and more thermally stable magnetic grain and so more capacity per platter. The important consequence for a buyer is what it does not do: unlike shingled recording, it imposes no rewrite penalty. A HAMR drive behaves like a conventional drive from the operating system's point of view, which is why it can go into an array where a shingled drive cannot.

Seagate's implementation is branded Mozaic. The 3+ generation underpins the 28, 30 and 32 TB drives, built on ten platters, with sequential transfer around 275 MB/s and a rated 2.5 million hour MTBF. Our MTBF to AFR calculator converts that rating into the annualized figure it implies, which is a more useful way to read it. The 36 TB part sampled in January 2025 has not been confirmed at retail, and the 44 TB Mozaic 4+ announced in March 2026 went to hyperscale customers rather than shelves.

Western Digital took a different route to high capacity. Its 40 TB drive, shipping since August 2026, uses UltraSMR, a shingled format that overlaps tracks to gain density and pays for it on rewrite. For an array that is a meaningful limitation: rebuilds and sustained random writes can slow dramatically. WD's practical maximum for array use is therefore the 26 TB conventional drive, not the 40 TB headline, with its HAMR part stated for the first half of 2027. Toshiba has stayed with microwave-assisted recording in its current high-capacity lines and has indicated HAMR around 2027.

None of this has reduced the price per terabyte at retail, and it is worth being clear about why. Higher areal density lowers the cost of manufacturing a terabyte, but during a shortage the highest-capacity output goes to the customers who contracted for it first. Seagate has described the vast majority of nearline exabytes as allocated into calendar 2028. What reaches a consumer shelf is the previous generation, priced against scarcity.

Who it hits, and how hard

Array builders should read the recording column before the capacity column. A 40 TB shingled drive and a 26 TB conventional one are not the same product for RAID or ZFS, and the larger number is the worse choice for that use.

Buyers waiting for a bigger drive should note how little of the roadmap reaches retail. The 36 TB and 44 TB parts are sampled or hyperscale-only. The largest drive you can actually buy today is 30 TB, and it has been available since July 2025.

Anyone expecting HAMR to cut prices should set that expectation aside for now. No published forecast ties HAMR volume to retail price relief, and the allocation statements point the other way for as long as the shortage lasts.

Anyone worried about reliability of a new recording technology is asking a reasonable question with no good public answer yet. These drives are too new to appear meaningfully in the Backblaze dataset, so anyone offering a verdict either way is ahead of the evidence. Our failure rate table will show them when the data exists.

What to do now

1. Check recording technology before capacity with our model number decoder; a shingled drive is a poor fit for an array at any size.
2. Buy the largest conventional drive your rebuild window tolerates, not the largest drive on a roadmap.
3. Convert any MTBF you are quoted into an annualized rate with our MTBF calculator before treating it as a lifetime.
4. Compare the per-terabyte cost of the biggest retail drives against mid-range ones on our price per TB page; the largest is not always the cheapest per terabyte.
5. Do not wait for a 44 TB drive. It is not a retail product and nobody has said when it will be.
6. If the reason you want more capacity is cost per terabyte, check why prices rose first; the answer is supply allocation, which a bigger drive does not solve.

Outlook

The dated statements on record run to the first half of 2027 for Western Digital's HAMR part and around 2027 for Toshiba's. Seagate is already shipping and has announced through 44 TB for hyperscale. Nothing in that record promises a retail price effect, and the allocation commitments argue against one while they are being filled.

We publish no forecast of our own. Everything above is someone else's published position, with the date it was published and a link to it, because a price forecast from a site that sells against those prices is worth nothing to you.