Backblaze's Q2 2026 Failure Rate Hit 1.73%, and Three Old Models Explain Much of It
Hardware / explainer
Backblaze's Q2 2026 Failure Rate Hit 1.73%, and Three Old Models Explain Much of It
The three outlier models are 3.4% of 354,415 drives and about 15% of failures. A third-party tally differs by one failure.
Backblaze's quarterly failure rate for hard drives reached 1.73% in the second quarter of 2026, which the company's own report calls "the highest it's been in quite a while," and 3 old drive models account for much of the movement. They are 3.4% of the drives Backblaze analysed and about 15% of its failures, by this piece's arithmetic from the report's table.
Stephanie Doyle, Backblaze's technical narrative content manager, wrote the Q2 2026 Drive Stats report, published Sept. 29 and covering April 1 to June 30. It is operator data from one company's data centres, not a lab test, and its published dataset is what third parties, including the analysis cited below, query.
What counts as a Backblaze annualised failure rate
Backblaze monitored 359,101 drives in the quarter. It removed 3,881 boot drives and 705 drives that did not meet the report's criteria, leaving 354,415. Those drives logged 31,553,350 drive days in the quarter and 558,148,607 over their lifetimes. Annualised failure rate, or AFR, is failures divided by drive days, multiplied by 365.
At 1.73% the quarterly rate implies roughly 1,500 failures across the fleet; that is this piece's calculation, since the report gives rates and the table gives per-model failure counts. The lifetime AFR ticked up to 1.41%. The report says 10 of 31 models had an AFR above 3.0%, a distribution "skewed towards the higher end of the spectrum" that "will, of course, bring up our average."
The three outliers above 6.95%
Three models crossed the report's outlier threshold of 6.95%.
| Model | Drives | Average age | AFR |
|---|---|---|---|
| Seagate ST10000NM0086 (10TB) | 965 | 101.7 months | 9.33% |
| Seagate ST14000NM0138 (14TB) | 1,235 | 65.5 months | 8.26% |
| HGST HUH721212ALN604 (12TB) | 9,694 | 83.8 months | 7.63% |
Together the three are 11,894 drives and 227 failures. The report attributes the numbers to age and small populations. The HGST model is the one that matters for the average: 9,694 drives, 861,297 drive days and 180 failures. The two Seagate models together contribute 47 failures on 2,200 drives, a sample small enough that a few drives either way moves the percentage by a point.
Where the bulk of the fleet sits
The three largest populations in the report look nothing like the outliers. Western Digital's WUH722222ALE6L4 (22TB) has 45,665 drives at 0.55%. Seagate's ST16000NM001G (16TB) has 35,100 at 0.65%. Toshiba's MG07ACA14TA (14TB) has 37,171 at 1.94%. Three Seagate models had no failures in the quarter: the 8TB ST8000NM000A, the 12TB ST12000NM000J and the 14TB ST14000NM000J.
- WDC 22TB (45,665 drives)0.55 %
- Seagate 16TB (35,100 drives)0.65 %
- Toshiba 14TB (37,171 drives)1.94 %
- HGST 12TB (9,694 drives)7.63 %
- Seagate 14TB (1,235 drives)8.26 %
- Seagate 10TB (965 drives)9.33 %
Source: Backblaze Q2 2026 Drive Stats report, published Sept. 29, 2026
The newest big model is not free of questions. The 24TB Seagate ST24000NM002H posted a 3.3% quarterly AFR and 3.24% lifetime in Backblaze's table. Both figures sit above the 3.0% line the report uses when counting how many models skew high.
When two tallies of the same data disagree
An independent analysis of Backblaze's public dataset, published by The Average Guy using DuckDB and Apache Iceberg queries, matches Backblaze on the HGST (7.628%) and Seagate 14TB (8.265%) models. It differs on two others. It gives the Seagate 10TB model 9.751%, against Backblaze's 9.33%, and the Seagate 24TB model 3.474%, against 3.3%.
The 10TB gap is checkable. 22 failures over 86,092 drive days is 9.33%; 23 failures over the same days is 9.75%. So the two tallies differ by one failure on that model, which may reflect a drive the report excluded. Neither source explains it. The author also cautions that data-centre conditions differ from home use, and the same limit applies here.
How to read it, with a link to scale
Backblaze's rate is a fleet statistic, and fleets are exactly where cost lives: a 1.73% rate across 354,415 drives is a replacement schedule, not a probability a single buyer will meet. The denominator for a home user is one drive, and the report says model, age and sample size matter more than the label on the box. Our piece on Microsoft's Titan flaw and its 17 trillion rows is about the scale of data that sits on drives like these, and Digit 5's unit economics makes the same point about denominators in another market.
The photograph above shows a 2011 Backblaze pod, not a current one. What would change this read is the Q3 report: if the three old models are retired and the rate stays above 1.7%, the cause is not just age.
Sources
More in Hardware
- 01Rio Tinto and Rondo Start Design of a 360 MWh Heat Battery at Yarwun RefineryA 30 MW discharge rating and a 4 to 6 hour solar window imply charging power well above 30 MW. The study is funded; construction is not.
- 02Porsche Wireless Charging: 90 Percent Efficiency Is the Number to WatchAn 11 kW floor pad for the Cayenne Electric is orderable in Europe, and the five-point efficiency gap to a cable adds up to roughly 4 kWh per fast home top-up.
- 03Tesla Dates Unsupervised FSD for Owners at Q2 2027, Inside Robotaxi Zones OnlyAshok Elluswamy gave the date on the Q3 call, and Electrek reports the first version is a way to add owned cars to Tesla's managed Robotaxi fleet.
- 04Tenstorrent Raises $1.5 Billion at $6.2 Billion, 24 Percent Above Reported Takeover InterestNeither TechCrunch nor BetaKit reports a Galaxy server spec, price, customer or revenue figure for the chip maker's Series E.