Ultrastar Deals - New and Refurb Verdicts, Live

Reading a dual-track verdict

Every row above compares a drive against its own tracked history — and crucially, against its own history on its own condition track. A drive's refurbished price series and its new price series are different series describing different products. Blending them would produce a verdict that describes neither, and it would erase the single most useful thing this page can show you.

That thing is the crossover. A refurbished unit sitting at its tracked low next to a new unit of the same capacity reading ELEVATED is the moment the pull market is unambiguously the right route, and it is invisible on any page that treats condition as a footnote. The reverse happens too: a new drive at its low beside a refurbished one that has drifted up, which is the week to buy new.

The thresholds are the same on both tracks. Within three percent of the thirty-day minimum reads AT/NEAR LOW; within seven percent either side of the median reads TYPICAL; more than seven percent above reads ELEVATED. Applying identical rules to both tracks is what makes them comparable at a glance.

NEW means fewer than seven days of history on that track, and it is not a judgement. A drive we started tracking this week has a real price and no meaningful range, so we show the price and the date tracking began rather than manufacturing a thirty-day window from a few hours of data. On a freshly seeded line most rows will read this way at first, and they resolve into real verdicts as history accrues.

Why refurbished Ultrastar is the deal-hunter's default

Ultrastar is the flagship of the used-enterprise market, and the reason is structural rather than sentimental. Datacenter operators replace drives on a schedule rather than at failure, so a steady supply of working, tested, high-capacity drives flows into resale every year. The drives were built for continuous operation, they were run in controlled environments, and they were retired while still functional.

In a normal market that supply would compete with cheap new drives and the discount would be modest. In 2026 it competes with new drives priced into a documented shortage, which widens the gap considerably at some capacities. The ladder on the Ultrastar hub shows the current spread per capacity, and the wide ones are why deal communities treat pulls as the default rather than the compromise.

What you are trading away is the manufacturer warranty and the certainty of an unused drive. What you are getting is enterprise-class hardware, frequently at a fraction of the new price, from a market that exists precisely because these drives outlast their deployment schedules.

The verdicts above tell you whether today is a good day on that track. The hub ladder tells you whether the track itself is worth taking at your capacity. Both are needed: a refurbished drive at its own tracked low is not a good deal if the new price at that capacity has collapsed to nearly the same number that week.

When new wins anyway

Three situations make the new track correct regardless of what the spread says, and it is worth naming them because the refurb-first instinct can be expensive in exactly these cases.

The first is warranty-critical deployment. If the drive is going into something where a failure means a support contract, a client obligation or a compliance question, a manufacturer warranty is a different instrument from a seller warranty, and the difference is the whole point. Our refurbished warranty guide covers what each actually obliges.

The second is a single-drive role holding data that is not redundantly stored. A pull in an array with parity is a managed risk; the same pull as the only copy of something is not, and no discount changes that arithmetic.

The third is simply appetite. Buying refurbished properly means verifying SMART on arrival, knowing what power-on hours you are looking at, and being willing to return a drive that arrives wrong. That is a small amount of work, but it is not zero, and someone who will not do it should not buy into the pull market. Paying the new-drive premium to skip that step is a legitimate choice rather than a failure of nerve.

For everyone else — arrays with redundancy, replaceable data, and a willingness to spend ten minutes with a SMART readout — the spread on the hub ladder is usually decisive.

The false-comfort trap

The most expensive mistake in this market is not buying a high-hours drive. It is buying a drive whose history you cannot see.

Some sellers clear SMART data before relisting. The stated reason is presentation; the effect is that the drive's power-on hours, reallocated sectors and error counters all read as though it were nearly new. A buyer comparing two listings — one disclosing forty thousand hours, one showing almost none — will reach for the second, and may be buying the older drive.

So the guidance is the reverse of the instinct. A seller who discloses a high hour count is giving you information and pricing accordingly. A drive whose counters look implausibly clean for a datacenter pull is giving you nothing, and the absence is the warning. We are not going to attach figures to how common this is, because we have not measured it and nobody credible has.

What you can do is concrete. Check SMART on arrival before writing anything important. Compare the power-on hours against what the listing implied. Look at reallocated and pending sector counts rather than just the hour meter. And treat a drive that arrives with a suspiciously blank history as a return candidate rather than a bargain. The recertified drives page carries the full checklist, and drive health covers reading the attributes that matter.

One more physical check belongs here: if a pull arrives and appears completely dead, suspect the SATA 3.3V power-disable pin before suspecting the drive. Our shucking guide covers it, and it accounts for a meaningful share of dead-on-arrival reports that are not dead at all.

What the tracked-since date is telling you

Every row we cannot yet judge carries the date we started watching it, and that date is more useful than it looks. It tells you how much weight to put on the verdict beside it once one appears.

A drive at seven days of history has a technically valid thirty-day window built from a week of observations. It is real data and it is thin. The same drive at thirty days has a window worth acting on. We do not grade that distinction with a confidence score, because a score would imply a precision we do not have — the date is the honest version, and it lets you weight the verdict yourself.

This matters most on a freshly expanded line. When we add models, their rows read NEW for the first week regardless of how good their prices are, and a page full of NEW badges can look like a page with nothing to say. It is the opposite: it is the page declining to say something it cannot support yet.

The corollary is that the verdicts on long-tracked drives are the ones to lean on hardest. A model we have watched for months has a range shaped by real promotions, real shortages and real recoveries, and a verdict against that range means something a week-old series cannot.

Capacity jumps on two tracks

The usual capacity-ladder advice applies here with an extra dimension. Cost per terabyte does not scale linearly, so the rung above the one you planned is frequently better value — but on this line you have to check that on the track you are actually buying.

The new and refurbished ladders do not have the same shape. Refurb supply depends on what datacenters retired, which is driven by deployment cycles from years ago rather than by current manufacturing. That produces capacities where the pull market is deep and cheap while the new market is tight, and others where the reverse holds. A capacity jump that improves your rate on the new track can worsen it on the refurbished one.

The practical check takes a moment: look at your intended capacity and the one above it on the hub ladder, on the track you intend to buy, and compare both the per-terabyte rate and the absolute outlay. If the rate improves and the extra money is tolerable, take the larger drive.

The array caveat still applies. In a parity array, mixing capacities usually wastes the extra space on the larger drives until every drive is replaced, so a good deal on a bigger drive may buy you nothing immediately. Our upgrade planner models that properly.

Frequently asked questions

Sources

Specification claims on this page were checked against the pages below on 4 August 2026. Where a widely-repeated figure did not survive that check, we report what the source says now and say so in the text rather than repeating the familiar number.

1. Western Digital — Ultrastar data center drives (westerndigital.com, 2026)
   Used for: product line and series naming. WD's datasheet PDFs and product pages did not yield workload rate or warranty term through our fetch path, which is why both columns render dashes rather than a quoted figure.
2. DatacenterDisk Research — Per-part-number interface decoding (DatacenterDisk.com, Live)
   Used for: interface per row taken from the listing title and cross-checked against the part-number suffix (ALE/ALN/ALA/BLE = SATA; AL5/AL4/ALS/CSS = SAS). Candidates whose title and part number disagreed were rejected; the pattern was validated against all 25 Ultrastar part numbers already in the catalogue with zero disagreements.
3. Backblaze — Published hard drive fleet statistics (backblaze.com, 2024-2026)
   Used for: directional only — HGST and Ultrastar models have historically appeared among lower-failure-rate entries in Backblaze's published fleet stats. No percentages or per-model rates are taken from it.
4. DatacenterDisk Research — Live Ultrastar price and stock database, both condition tracks (DatacenterDisk.com, Live)
   Used for: every price, $/TB figure, condition track, stock state and price-history range on these pages
5. DatacenterDisk Research — Recertified drives — verify-on-arrival checklist (DatacenterDisk.com, 2026)
   Used for: refurbished-purchase guidance referenced rather than duplicated
6. DatacenterDisk Research — Shucking — the 3.3V power-disable pin (DatacenterDisk.com, 2026)
   Used for: the pin behaviour behind drives that appear dead on arrival