Outage frequency per site has now fallen for a fifth consecutive year, while the share of outages costing more than USD 100,000 rose from 54% to 57%. Those two trends running in opposite directions are the defining feature of IT reliability data right now: infrastructure is failing less often but each failure is more expensive when it happens. One in five organisations report their most recent impactful outage cost more than USD 1 million. The figures below come from the Uptime Institute Annual Outage Analysis, now in its eighth annual edition.
TL;DR
- Outage frequency per site has declined for a fifth consecutive year (Uptime Institute)
- The rate of improvement has slowed (Uptime Institute)
- 57% say their most recent major outage cost more than USD 100,000 (Uptime Institute, 2025 survey)
- That is up from 54% in the 2024 survey (Uptime Institute)
- One in five report an outage costing more than USD 1 million (Uptime Institute)
- Roughly 1 in 10 report serious or severe impact from their last outage (Uptime Institute)
- Power is the leading cause (Uptime Institute)
- Specifically UPS systems, transfer switches, and generators (Uptime Institute)
- Major data center fires have increased gradually (Uptime Institute)
- Lithium-ion UPS batteries are a clear contributing factor (Uptime Institute)
- External infrastructure failures are more prominent in publicly reported outages (Uptime Institute)
- Many AI sites are not directly covered by the report (Uptime Institute)
- The 2026 edition is the eighth annual report (Uptime Institute)
1. Frequency Is Falling, Slowly
The headline trend is genuinely good news and has been for half a decade. Uptime Intelligence research indicates that outage frequency on a per-site basis has declined for a fifth consecutive year, a run long enough to rule out noise.
The qualifier matters as much as the trend. Uptime notes that the rate of improvement has slowed compared with previous years, which is the pattern you would expect from an industry harvesting the easy gains first. Redundancy design, monitoring, and operational procedure have improved substantially, and each additional increment of reliability now costs more than the last. A decelerating improvement curve against a backdrop of rising complexity is a reasonable description of a sector approaching the limits of its current approach rather than one that has solved the problem.
| Metric | Value | Source |
|---|---|---|
| Consecutive years of declining per-site frequency | 5 | Uptime Institute |
| Direction of the trend | improving | Uptime Institute |
| Rate of improvement | slowing | Uptime Institute |
| Measurement basis | per site | Uptime Institute |
| Report edition | 8th annual | Uptime Institute |
| Publication month | May 2026 | Uptime Institute |
| Full report length | 29 pages | Uptime Institute |
| Executive summary length | 9 pages | Uptime Institute |
Source: Uptime Institute Annual Data Center Outages Analysis 2026.
2. Cost Is Moving the Other Way
While incidents get rarer, they get more expensive, and the cost distribution has shifted measurably in a single year. 57% of respondents to Uptime’s 2025 annual survey said their most recent major outage cost more than USD 100,000, up from 54% in the 2024 survey.
A three-point shift in one year is modest in isolation but consequential in direction, because it means the majority threshold is moving further into expensive territory. Roughly 43% of outages now fall below USD 100,000 against 46% a year earlier. The plausible mechanism is consolidation: as workloads concentrate onto fewer, larger, denser sites, each failure touches more services and more revenue. That is the same structural change that drives the frequency improvement, since fewer and better-run facilities fail less often, which means the two trends are likely two faces of one process rather than independent developments.
| Metric | Value | Source |
|---|---|---|
| Outages costing more than USD 100,000, 2025 survey | 57% | Uptime Institute |
| Outages costing more than USD 100,000, 2024 survey | 54% | Uptime Institute |
| Year-over-year change | up 3 points | Derived from Uptime figures |
| Outages below USD 100,000, 2025 survey | approx. 43% | Derived from Uptime figures |
| Outages below USD 100,000, 2024 survey | approx. 46% | Derived from Uptime figures |
| Outages costing more than USD 1 million | 1 in 5 | Uptime Institute |
| Duration of the USD 1 million finding | reported in consecutive editions | Uptime Institute |
| Direction of cost trend | rising | Uptime Institute |
Infrastructure context sits in our data center statistics. Source: Uptime Institute 2025 outage analysis press release.
3. The Severity Distribution
Aggregate cost figures obscure how lopsided the severity distribution is. Roughly 1 in 10 impacted organisations report that their last outage had serious or severe consequences, which means around 9 in 10 fell below that threshold.
This is the number that should temper alarmist readings of outage data. The typical outage is an operational nuisance, not a business-threatening event, and most organisations experiencing one absorb it without lasting consequence. The risk lives entirely in the tail. Combining this with the cost data gives a coherent picture: a large majority of incidents are minor, a fifth cross USD 1 million, and about a tenth do serious damage. Planning against the median outage and planning against the tail are different exercises, and only the second one justifies the cost of deep resilience investment.
| Metric | Value | Source |
|---|---|---|
| Serious or severe impact, last outage | approx. 1 in 10 | Uptime Institute |
| Below that threshold | approx. 9 in 10 | Derived from Uptime figures |
| Outages above USD 1 million | 1 in 5 | Uptime Institute |
| Outages above USD 100,000 | 57% | Uptime Institute |
| Shape of the distribution | heavy tail, minor median | Derived |
| What most outages are | operational nuisance | Derived |
| Where the risk concentrates | the tail | Derived |
| Planning implication | tail risk drives resilience spend | Derived |
Source: Data Center Knowledge on the 2026 outage report.
4. Power Remains the Leading Cause
Across editions of this analysis one category keeps coming first. Uptime identifies power issues as the leading cause of outages, specifically failures involving uninterruptible power supply systems, transfer switches, and generators.
The consistency is itself the finding. Power has topped this ranking year after year despite being the best-understood failure domain in the facility, with mature redundancy standards and decades of engineering practice behind it. That persistence suggests the problem is not knowledge but execution at scale: UPS systems, transfer switches, and generators are electromechanical components with maintenance cycles and failure modes that no amount of software sophistication removes. Uptime also flags grid instability as a rising external pressure, which pushes more load onto exactly these components at exactly the moment they are most likely to be tested.
| Metric | Value | Source |
|---|---|---|
| Leading cause category | power | Uptime Institute |
| Specific components implicated | UPS systems, transfer switches, generators | Uptime Institute |
| Consistency across editions | leading cause repeatedly | Uptime Institute |
| Nature of the components | electromechanical | Derived |
| Rising external pressure | grid instability | Uptime Institute |
| Other pressures identified | demand growth, AI workloads, power constraints | Uptime Institute |
| Additional factors | system complexity, deeper interdependencies | Uptime Institute |
| External infrastructure failures | more prominent in publicly reported outages | Uptime Institute |
Cloud dependency context sits in our cloud computing statistics. Source: Uptime Institute Annual Data Center Outages Analysis 2026.
5. Fires Are Trending the Wrong Way
Most risk categories in this dataset are improving. One is not. Uptime reports that the frequency of major fires at data centers has increased gradually in recent years, with lithium-ion batteries in UPS systems a clear contributing factor.
The causal chain here is unusually legible. Operators moved from valve-regulated lead-acid to lithium-ion UPS batteries for good reasons, including higher energy density, longer service life, smaller footprint, and faster recharge. Lithium-ion chemistry also carries thermal runaway risk that lead-acid does not, and concentrating more energy in less space raises the consequence of any single cell failure. This is a straightforward engineering trade-off that the industry accepted, and the fire trend is the cost side of it becoming visible in the data. It is also the clearest example in this report of a resilience improvement in one dimension creating exposure in another.
| Metric | Value | Source |
|---|---|---|
| Trend in major data center fires | increasing gradually | Uptime Institute |
| Identified contributing factor | lithium-ion UPS batteries | Uptime Institute |
| Strength of the attribution | clear, per Uptime | Uptime Institute |
| Prior battery technology | valve-regulated lead-acid | Derived |
| Reasons for the transition | energy density, service life, footprint | Derived |
| Risk introduced | thermal runaway | Derived |
| Direction versus other categories | opposite, worsening | Derived |
| Nature of the finding | trade-off cost becoming visible | Derived |
Security-side incident context sits in our cybersecurity statistics. Source: Uptime announces the 2026 outage analysis report.
6. The AI Blind Spot
The most important limitation is one the report states about itself. Uptime notes that many AI sites are still relatively new and are therefore not directly covered by the report, even as it identifies AI-driven workloads as a factor reshaping risk profiles.
That gap is awkward and worth sitting with. The analysis names demand growth, AI workloads, and power constraints as the forces changing the risk landscape, then measures a population that largely predates those forces. AI training and inference facilities differ from conventional enterprise sites in power density, cooling approach, and utilisation pattern, all of which bear directly on failure modes. The reliability improvements documented here are real for the infrastructure they describe, but they should not be read as a forecast for the buildout now underway. The next several editions will be considerably more informative on that question than this one can be.
| Metric | Value | Source |
|---|---|---|
| Coverage of AI sites | largely not directly covered | Uptime Institute |
| Stated reason | many AI sites are relatively new | Uptime Institute |
| AI named as a risk factor | yes | Uptime Institute |
| Data sources used | public reports, surveys, members and partners, outage database | Uptime Institute |
| Report authors | Douglas Donnellan, Andy Lawrence, Rose Weinschenk | Uptime Institute |
| Edition | 8th annual | Uptime Institute |
| Publication | May 2026 | Uptime Institute |
| Reliable inference for AI facilities | limited | Derived |
Source: Uptime Intelligence annual outage analysis 2026.
Summary: IT Outages by the Numbers
| Metric | Value | Source |
|---|---|---|
| Consecutive years of declining per-site frequency | 5 | Uptime Institute |
| Rate of improvement | slowing | Uptime Institute |
| Outages above USD 100,000, 2025 survey | 57% | Uptime Institute |
| Outages above USD 100,000, 2024 survey | 54% | Uptime Institute |
| Year-over-year change | up 3 points | Derived |
| Outages above USD 1 million | 1 in 5 | Uptime Institute |
| Serious or severe impact | approx. 1 in 10 | Uptime Institute |
| Below serious threshold | approx. 9 in 10 | Derived |
| Leading cause | power | Uptime Institute |
| Components implicated | UPS, transfer switches, generators | Uptime Institute |
| Major fire trend | increasing gradually | Uptime Institute |
| Fire contributing factor | lithium-ion UPS batteries | Uptime Institute |
| External infrastructure failures | more prominent | Uptime Institute |
| AI site coverage | largely absent | Uptime Institute |
| Report edition | 8th annual | Uptime Institute |
| Publication | May 2026 | Uptime Institute |
Methodology and Sources
- Outage frequency trends, severity distribution, cause attribution, fire trends, and AI coverage limitations come from the Uptime Institute Annual Data Center Outages Analysis 2026, the eighth annual edition, published May 2026 as a 29-page report with a 9-page executive summary (Uptime Institute, Uptime Intelligence).
- The USD 100,000 cost thresholds for the 2024 and 2025 annual surveys come from Uptime’s own press releases for the corresponding report years (Uptime Institute 2025 press release, BusinessWire).
- Independent coverage was used to corroborate figures reported from the gated full report (Data Center Knowledge, Data Centre Solutions, PacketFabric).
- Data watch: the full report sits behind registration, so several figures here come from Uptime’s own press material and from press coverage of the report rather than from the primary document, and detailed cause breakdowns by percentage were not publicly available at the time of writing. The finding that one in five outages exceed USD 1 million is described as holding for a second consecutive year in more than one edition, which makes the exact survey year behind that specific figure ambiguous; it is presented here without a year attached for that reason. Cost figures are self-reported by survey respondents and are not independently audited. Frequency is measured per site, so a declining per-site rate is compatible with a rising absolute number of outages as the installed base grows. Rows marked as derived are arithmetic or direct inference from published figures.
- Last updated: August 2, 2026. We update this roundup quarterly, and the next major refresh is expected when Uptime Institute publishes its next annual outage analysis.