Backblaze Drive Stats 2025: Which Hard Drives Failed Least?

What changed and why

Backblaze publishes raw daily snapshots, one row per drive per day, and no summary dataset. We streamed every daily file for the four quarters of 2025 and counted, per model, the number of rows (drive days), the number of rows flagged as a failure, and the fleet size on the last day of the period. The rate is then failures divided by drive days times 365, expressed as a percentage. We do not copy the published figure, so the arithmetic above can be checked against the same public files.

Models below 250 drives or 50,000 drive days are marked: at that sample size one extra failure moves the rate by whole percentage points, so the number is not precise enough to rank on.

Has it moved since the annual report?

The 2025 annual figure is the headline, but quarters published since it are ingested the same way and computed with the same arithmetic. Each quarterly rate covers three months, so it moves around far more than an annual figure does.

What a low AFR does and does not tell you

These drives run in a datacentre: constant temperature, constant power, steady workload, and replaced on a schedule. That is gentler than a drive in a desk-side enclosure that is power-cycled daily and harsher than one that idles most of the week. The rates rank models against each other under one consistent regime, which is exactly what makes them useful, and they are not a prediction for your drive.

Who it hits, and how hard

Home NAS owners should read the ranking rather than the absolute rate. A model that fails more in a datacentre will not fail less in a cupboard, but your power cycles, temperature range and workload are all different, so the absolute number will not transfer.

Anyone sizing an array should treat the rate as an input to redundancy, not a substitute for it. Our MTBF to AFR calculator turns a rate into expected failures across a given number of drives.

Buyers comparing two models should compare the confidence ranges, not the headline rates. Where the ranges overlap heavily, the data cannot separate the drives and you should decide on warranty, workload rating and price instead.

Anyone shopping on reliability alone is probably optimising the wrong variable. The difference between a good drive and a bad deployment is larger than the difference between two well-behaved models, and backups matter more than either.

What to do now

1. Look up the exact model you are considering in our rate by model table, not the brand.
2. Check the drive days column before trusting a rate; a small sample is not a result.
3. Convert any rated MTBF on a datasheet into an annualized figure with our MTBF calculator.
4. Confirm the recording technology with our model number decoder before putting a drive in an array.
5. Compare makers only after reading the per-model table, using our brand comparison and its caveats.
6. Keep backups regardless. RAID is not a backup and no failure rate changes that.

Outlook

Backblaze publishes quarterly, with an annual report early in the following year. We re-ingest when a new period appears and recompute rather than copying the published figure, so the method stays constant across periods. The quarterly table above shows whether the fleet rate has moved since the annual report.

We make no prediction about future failure rates. A drive model's rate in one year is a weak predictor of the next, because the population ages and the fleet mix changes.