Seagate vs Western Digital vs Toshiba: Which Drives Fail Least?

What changed and why

Brands we will not rank

A brand needs a real population before its rate means anything. Every maker in this period cleared our floor of 500,000 drive days. A zero percent rate over a few drive-years is an absence of evidence, not a result, and leading a brand comparison with one would be the exact error the rest of this page warns about.

Why these are weighted by drive days

Averaging each maker's model rates would let a model with 300 drives count as much as one with 60,000. That single choice is how the same dataset gets used to argue opposite conclusions about the same brands. Weighting by drive days asks a different and more answerable question: across all the running time Backblaze logged for this maker, what share of drives failed per year.

Boot devices are not counted

Backblaze runs boot SSDs and small laptop drives alongside its data fleet. They are not hard drives and they are not what this page is about, so everything below 4 TB is excluded. That exclusion is what makes our fleet figure reproduce Backblaze's published rate: including the boot devices only dilutes the denominator, because none of the year's failures were theirs.

What this comparison cannot tell you

Backblaze does not buy an even spread of models across makers. In any given year one maker may be represented mostly by a recent, well-behaved high-capacity line and another by an aging deployment bought years earlier. The brand totals carry that mix with them, and no weighting fixes it.

Treat the per-model table as the real result and the brand totals as context. A maker with a good overall figure still sells models with poor rates, and the model number is what you actually buy. Our rate by model page has that breakdown.

Who it hits, and how hard

Anyone shopping by brand is using the weakest signal available here. The spread inside each maker's range is wider than the gap between makers, and you buy a model number.

Fleet operators buying in volume get more from this view, because they are genuinely exposed to a maker's whole range rather than to one drive.

Anyone citing a brand verdict from this or any other dataset should state the weighting. Averaging model rates instead of weighting by drive days produces a different answer from the same numbers, which is how this dataset ends up supporting opposite conclusions.

Owners of an HGST drive should note that those are a distinct and generally older population rather than current Western Digital products, which is why they are listed separately.

What to do now

1. Start from the per-model table, not this one; the model is what you buy.
2. Check the drive days column here before treating any brand figure as meaningful.
3. Read how the rates are computed on our 2025 Drive Stats page.
4. Confirm recording technology for the specific model with the decoder.
5. Size redundancy from the rate with the MTBF to AFR calculator.
6. Ignore brand loyalty arguments that do not state a weighting and a sample size.

Outlook

Brand composition changes as Backblaze buys. A maker that looks good in one year may be represented mostly by a recent, well-behaved line, and a maker that looks poor may be carrying an aging deployment bought years earlier. No weighting fixes that, so treat the brand totals as context and the per-model table as the result.

We make no prediction about future rates by maker.