In a time when technological discussions often revolve around Artificial Intelligence and chips, Backblaze has once again done something less glamorous but incredibly useful for any operations team: publishing data. Their annual report Drive Stats 2025 marks 13 years of continuous monitoring of hard drive performance in production—a series that, over time, has become an uncommon thermometer for understanding what really happens when hundreds of thousands of units spin day and night in data centers.
This year’s headline is clear: the annualized failure rate (AFR) for 2025 drops to 1.36%, below the 1.55% of 2024. In the last quarter (Q4 2025), the figure improves even further to 1.13%, while the “lifetime” view remains almost steady at 1.30%. In industry terms: the disk fleet appears healthier, but maintenance work doesn’t disappear; it simply changes form.
An impressive sample: hundreds of thousands of disks under watch
Backblaze explains that, at the end of 2025, it was monitoring 349,462 storage units for its systems. For the annual analysis, it excludes boot drives and others that don’t meet its statistical criteria, ending up with 344,196 valid disks spread across 30 models. The report doesn’t just “count failures”: it also contextualizes by age, “disk days,” and operational changes, to avoid misleading conclusions when the sample size is small or the hardware is at the end of its lifespan.
This approach is key to interpreting the numbers correctly: in real environments, the “how” is as important as the “how many.” A spike in failures could be due to firmware defect, yes, but also persistent vibration, an anomalous rack, a specific batch, or simply wear and tear after years of service.
What does AFR mean and why does Backblaze set “limits” on its own dataset
The AFR (Annualized Failure Rate) is a way to annualize failures for comparison over time, even when analyzing a quarter. Additionally, Backblaze sets minimum thresholds for including models in tables and charts: for example, in quarterly analysis, more than 100 units and over 10,000 “drive days”; annually, more than 250 units and over 50,000 “drive days”; and in “lifetime” tables, more than 500 units and over 100,000 “drive days”. The goal is simple: prevent a few failures in a small group from skewing the overall picture.
For those purchasing or managing storage, the practical lesson is clear: reliability comparisons are only useful if the sample size and operational time are sufficient. And, even then, it’s important to look at trends, not just the snapshot data.
Larger capacity in fewer disks: the fleet shifts toward bigger sizes
One of the most interesting visuals in the report is the capacity distribution. Backblaze observes that storage weight is shifting toward large disks: in their breakdown, disks of 0–12 TB account for 25.13%, those of 14–16 TB rise to 52.06%, and disks of 20 TB or more make up 22.81%. This confirms a trend seen in many data centers: reducing the number of units needed to store the same amount (and thus lowering potential failure points and replacement work), although this requires more precise planning for capacity, rebuilds, and resilience policies.
Within this context, the report also highlights the arrival of new units in the fleet, including the introduction of a 26 TB model from Western Digital (their “first 26 TB drive” in production), indicating that the race for higher density continues to accelerate.
The “honor roll” and alerts: when the issue isn’t the disk but the environment
Backblaze complements its tables with a very human list: models that “perform well” and models that trigger alarms. In Q4 2025, three cases stand out due to their elevated quarterly AFR:
- HGST HUH728080ALE600 (8 TB): 10.29%
- Seagate ST10000NM0086 (10 TB): 5.23%
- Toshiba MG08ACA16TEY (16 TB): 4.14%
In the case of the 8 TB HGST, the company adds context that breaks the “bad model” cliché: the analyzed group consisted of 1,073 disks, fewer than a standard “Vault” of 1,200, suggesting physical concentration. Internal investigation ruled out temperature as the main cause and pointed instead to vibration sensitivity; nonetheless, age (around 7.5 years) is a factor, and operational decision was made to mark these units for normal migration and retirement.
The Toshiba MG08ACA16TEY case is also revealing for another reason: it illustrates the impact of firmware and mitigation strategies. After recording an exceptionally high AFR in the previous quarter, Backblaze mentions collaborative firmware work and a gradual normalization. Although the figure remains above ideal, the trend suggests that the issue isn’t always hardware as such but how its maturity is managed in production.
Operational reality: disk replacement remains a daily task
Even as the AFR improves, logistics don’t stop. Only in Q4 2025, 943 disks failed—that’s a cadence equivalent to dozens of replacements daily in a large fleet. This is the less glamorous side of mass storage: maintenance isn’t a “once-off event,” but an industrial routine.
Therefore, Drive Stats is especially relevant for system teams: not just to pick “the perfect disk,” but to refine policies on lifecycle management, spare parts, preventive migrations, vibration control, and observability tools (SMART, alerts, tickets) that prevent a local problem from becoming a widespread statistic.
Frequently Asked Questions
What is the annualized failure rate (AFR) and how is it interpreted in data centers?
The AFR estimates what percentage of disks would fail in a year if the observed failure rate during a period were maintained. It’s useful for comparing quarter to quarter and year to year but should be read alongside “drive days,” sample size, and disk age.
What does it mean that Backblaze lowered the AFR to 1.36% in 2025?
It indicates that, within their monitored fleet, overall reliability has improved over 2024. It doesn’t mean “almost no disks fail,” but that proportionally, the replacement rate is lower… though the absolute number of replacements still remains high.
Which models sent warning signals in Q4 2025 according to Backblaze Drive Stats?
The most notable for high quarterly AFR are HGST HUH728080ALE600 (8 TB), Seagate ST10000NM0086 (10 TB), and Toshiba MG08ACA16TEY (16 TB). Backblaze adds operational context (age, environment, firmware) before drawing conclusions.
Can this data be used to select disks for NAS or enterprise servers?
It can serve as a reference, but with caution: results depend on environment, usage, age, and sample size. The most valuable insights usually come from trends and lifecycle management, rather than seeking an absolute “winner.”
via: backblaze