Hard drive failure rates fell to 1.24% annualized in the first quarter of 2026, and the largest drives are the most reliable ones in the fleet. Drives of 20TB and above posted a 0.85% annualized failure rate across more than 86,000 units, comfortably below the fleet-wide figure, which contradicts the long-standing intuition that higher-density drives fail more. Meanwhile 92% of newly deployed drives now exceed 20TB, driven by AI workload storage demand. The figures below come from Backblaze Drive Stats, the longest-running public dataset on hard drive reliability at scale.
TL;DR
- Q1 2026 annualized failure rate was 1.24% (Backblaze)
- That is down from 1.42% in Q1 2025 (Backblaze)
- Full-year 2025 annualized failure rate was 1.36% (Backblaze)
- That is down from 1.55% in 2024 (Backblaze)
- Lifetime annualized failure rate held steady at 1.30% (Backblaze)
- Drives of 20TB and above posted 0.85% AFR (Backblaze)
- That covers more than 86,000 units (Backblaze)
- 92% of newly deployed drives now exceed 20TB (Backblaze)
- The 22TB WDC WUH722222ALE6L4 posted 0.38% AFR (Backblaze)
- That model has more than 45,000 units in the fleet (Backblaze)
- The 24TB Toshiba MG11ACA24TE recorded a fleet-best 0.22% AFR (Backblaze)
- The dataset now spans thirteen years (Backblaze)
- AI workloads are cited as the driver of high-capacity deployment (Backblaze)
1. Failure Rates Are Falling
Two independent measures moved the same direction, which is what makes this a trend rather than a quarter. The annualized failure rate fell to 1.24% in the first quarter of 2026 from 1.42% a year earlier, and the full-year 2025 rate fell to 1.36% from 1.55% in 2024. Both are meaningful declines on a metric that has historically moved slowly, and the lifetime figure of 1.30% barely budged because it averages thirteen years of hardware together.
| Metric | Value | Source |
|---|---|---|
| Annualized failure rate, Q1 2026 | 1.24% | Backblaze |
| Same measure, Q1 2025 | 1.42% | Backblaze |
| Year-over-year change | down 0.18 points | Derived from Backblaze figures |
| Annualized failure rate, full-year 2025 | 1.36% | Backblaze |
| Same measure, full-year 2024 | 1.55% | Backblaze |
| Year-over-year change | down 0.19 points | Derived from Backblaze figures |
| Lifetime annualized failure rate | 1.30% | Backblaze |
| Years of data in the series | 13 | Backblaze |
The lifetime figure sitting between the improving recent rates and the worse historical ones is exactly what a stable long-run average should do, and it is the number to use when estimating expected failures over a full hardware lifecycle rather than in the current year.
Scale matters for interpreting these percentages. A 1.24% annualized rate sounds small until it is applied to a fleet of hundreds of thousands of drives, at which point it describes thousands of individual failure events a year, each requiring detection, replacement, and a rebuild of the redundancy that protected the data. This is why storage operators care about tenths of a percentage point that would be invisible to a consumer: the difference between 1.42% and 1.24% across a large fleet is hundreds of service events annually, and the labour and risk attached to them. It is also why the direction of travel matters more than the level. A rate that is falling while the fleet grows means the absolute failure count may stay flat even as capacity expands. Source: Backblaze Q1 2026 Drive Stats.
2. Bigger Drives Fail Less
This is the finding most likely to surprise anyone who has been buying storage for a decade. Drives of 20TB and above posted a 0.85% annualized failure rate across more than 86,000 units, roughly a third below the 1.24% fleet-wide rate. The received wisdom that cramming more platters and higher areal density into a drive increases failure risk simply does not hold for the current generation.
There is a selection effect worth naming before treating this as a pure engineering result. High-capacity drives are the newest in any fleet, and failure rates follow a bathtub curve in which young drives fail less than old ones regardless of capacity. Some portion of the 0.85% advantage is therefore age rather than design. The effect is unlikely to explain all of it across a population above 86,000 units, but it does mean the honest reading is that new large drives are outperforming a fleet average that includes much older hardware, not that capacity itself improves reliability.
| Metric | Value | Source |
|---|---|---|
| AFR for drives 20TB and above | 0.85% | Backblaze |
| Units in that population | more than 86,000 | Backblaze |
| Fleet-wide AFR for comparison | 1.24% | Backblaze |
| Relative difference | roughly one third lower | Derived from Backblaze figures |
| Direction versus conventional expectation | opposite | Derived |
| Largest single model population | 22TB WDC, 45,000+ units | Backblaze |
| That model’s AFR | 0.38% | Backblaze |
| Fleet-best AFR in early deployment | 0.22% | Backblaze |
There is a confound worth naming: high-capacity drives are also the newest drives, and failure rates follow a bathtub curve where young hardware fails less than aging hardware. Some of the advantage is capacity and some is simply age, and this data cannot cleanly separate the two. Source: Backblaze Drive Stats for 2025.
3. The Fleet Is Consolidating on High Capacity
Purchasing behaviour has shifted almost completely. 92% of newly deployed drives now exceed 20TB, which Backblaze attributes to increasingly data-intensive AI workloads. At that rate of turnover, the sub-20TB portion of any large fleet becomes a legacy tail within a few refresh cycles.
| Metric | Value | Source |
|---|---|---|
| Share of new deployments above 20TB | 92% | Backblaze |
| Stated driver | data-intensive AI workloads | Backblaze |
| Units at 20TB and above | more than 86,000 | Backblaze |
| AFR for that population | 0.85% | Backblaze |
| Largest single model | 22TB WDC WUH722222ALE6L4 | Backblaze |
| Units of that model | more than 45,000 | Backblaze |
| Newest high performer | 24TB Toshiba MG11ACA24TE | Backblaze |
| Fleet-wide AFR | 1.24% | Backblaze |
The operational logic is straightforward: fewer, larger drives means less rack space, less power, fewer failure events to service, and fewer objects to manage per petabyte stored. There is a countervailing risk that the headline reliability figures do not capture. When a 24TB drive fails, the rebuild pulls far more data across the array than a 4TB failure did, and that rebuild window is precisely when a second failure becomes catastrophic. Higher per-drive capacity therefore lowers the frequency of failure events while raising the consequence of each one, which is a trade rather than a straightforward improvement. Operators compensate with wider erasure coding and faster rebuild paths, but the arithmetic is worth stating because a falling AFR alone can make a fleet look safer than its recovery characteristics justify. Data centre context sits in our data center statistics. Source: Backblaze 2025 Drive Stats report announcement.
4. Best Performers, With a Caveat About Sample Size
Individual model results are the most-quoted part of this dataset and the most easily misread. The 24TB Toshiba MG11ACA24TE recorded a fleet-best 0.22% annualized failure rate, but in early deployment, meaning a small population observed for a short time. The 22TB WDC model’s 0.38% across more than 45,000 units is a far more trustworthy number even though it looks worse.
| Metric | Value | Source |
|---|---|---|
| Toshiba MG11ACA24TE AFR | 0.22% | Backblaze |
| Deployment stage | early | Backblaze |
| WDC WUH722222ALE6L4 AFR | 0.38% | Backblaze |
| Units of that model | more than 45,000 | Backblaze |
| Which figure is more reliable | the WDC result | Derived |
| Reason | far larger sample and longer observation | Derived |
| Capacity of the Toshiba model | 24TB | Backblaze |
| Capacity of the WDC model | 22TB | Backblaze |
Backblaze itself flags low-population models in its reporting for exactly this reason, and a model with a few hundred drives observed for one quarter can post a 0.00% rate that means very little. Component supply context sits in our semiconductor industry statistics and memory pricing in our RAM and DRAM statistics. Source: Backblaze Drive Stats for Q3 2025.
5. What This Data Can and Cannot Support
Drive Stats is unusually valuable because almost nothing else like it exists publicly, and that scarcity makes it prone to over-application. These are drives in climate-controlled data centers, powered continuously, running consistent workloads, handled by professionals. A desktop drive that is power-cycled daily, occasionally bumped, and cooled by a single case fan faces entirely different stress.
| Metric | Value | Source |
|---|---|---|
| Operating environment | climate-controlled data center | Backblaze |
| Power state | continuously powered | Backblaze |
| Workload character | consistent enterprise use | Backblaze |
| Applicability to consumer use | limited, treat as a floor | Derived |
| Fleet-wide AFR | 1.24% | Backblaze |
| Lifetime AFR | 1.30% | Backblaze |
| Years of published data | 13 | Backblaze |
| Whether raw data is public | yes, downloadable | Backblaze |
The one thing that makes this dataset genuinely exceptional is that Backblaze publishes the underlying test data rather than only the summary, so anyone can recompute the figures rather than taking them on trust. That is rarer than the statistics themselves. Consumer hardware context sits in our PC gaming hardware statistics. Source: Backblaze hard drive test data.
Summary: Hard Drive Reliability by the Numbers
| Metric | Value | Source |
|---|---|---|
| Annualized failure rate, Q1 2026 | 1.24% | Backblaze |
| Same measure, Q1 2025 | 1.42% | Backblaze |
| Quarterly year-over-year change | down 0.18 points | Derived |
| Annual failure rate, 2025 | 1.36% | Backblaze |
| Annual failure rate, 2024 | 1.55% | Backblaze |
| Annual year-over-year change | down 0.19 points | Derived |
| Lifetime annualized failure rate | 1.30% | Backblaze |
| AFR for drives 20TB and above | 0.85% | Backblaze |
| Units in that population | more than 86,000 | Backblaze |
| Share of new deployments above 20TB | 92% | Backblaze |
| Stated driver of high-capacity shift | AI workloads | Backblaze |
| WDC 22TB model AFR | 0.38% | Backblaze |
| Units of that model | more than 45,000 | Backblaze |
| Toshiba 24TB model AFR | 0.22% | Backblaze |
| Deployment stage of that model | early | Backblaze |
| Years of published data | 13 | Backblaze |
| Underlying test data published | yes | Backblaze |
Methodology and Sources
- All failure rates, model-level results, fleet composition, and capacity data come from Backblaze Drive Stats, published quarterly and annually from the company’s operational storage fleet (Q1 2026 report, 2025 annual report, 2025 report announcement, Q3 2025, Q2 2025).
- The underlying per-drive test data is published by Backblaze and can be downloaded and recomputed independently (hard drive test data).
- Data watch: this is one company’s fleet, not an industry sample. Backblaze buys drives in bulk based on price and availability, so its model mix reflects its purchasing decisions rather than the market, and some models have populations too small to support a meaningful rate. The company flags low-population models itself, and single-quarter results for newly deployed drives, including the 0.22% fleet-best figure here, rest on short observation windows and should not be read as durable rankings. Drives run continuously in climate-controlled facilities under consistent enterprise workloads, so these rates are a floor rather than a prediction for consumer use. The apparent reliability advantage of 20TB-plus drives is partly a capacity effect and partly an age effect, since high-capacity drives are also the newest in the fleet, and this dataset cannot separate the two. Rows marked as derived are arithmetic on published figures.
- Last updated: August 2, 2026. We update this roundup quarterly as Backblaze publishes new Drive Stats reports.