Outage Report: AI Boom Threatens Years of Data Center Resiliency Gains

extract

Outage Report: AI Boom Threatens Years of Data Center Resiliency Gains New research from Uptime Institute shows data center outages continuing to decline, but AI facilities could buck the trend. An industry rooted in resiliency continues to uphold its core tenets, as data center outages decline for the fifth consecutive year, according to Uptime Institute’s latest annual research. The 2026 Data Center Outage Analysis, published this week by Uptime’s research unit, examines downtime across the industry using a combination of the tier certification organization’s own survey, press reports, company statements, and other sources. However, a recent AI-driven pivot that appears to prioritize performance over resilience raises questions about how long the overall positive outage trend will continue. Despite the progress made in improving resilience, outages remain a fact of life for the data center industry. Recent high-profile incidents include an outage at an AWS facility in Northern Virginia, reportedly caused by a cooling issue. Andy Lawrence, executive director of research for Uptime Institute Intelligence, told Data Center Knowledge that the report’s findings make sense when viewed through a multi‑year lens. “This year’s results do not show anything dramatic or unexpected, in the sense that the outage rate has, for the fifth year running, shown a small improvement,” he said. “But at the same time, the improvement has slowed, and there are signs that some new factors that could cause more outages are coming more into play. This ranges from increasing grid instability, more cyber threats, and the very noticeable trend of cable cuts.” External Factors Driving Outages Citing subsea cable cuts and cyber-attacks as notable examples, the latest Uptime data points to an increase in IT service outages being caused by external infrastructure rather than by failures within a facility. However, despite the escalation in anti-data center sentiment across the US and globally, Uptime’s Lawrence said acts of deliberate sabotage on facilities were still rare. Other key findings from the report include: Outage costs continue to increase: For the second year in a row, one in five respondents reported costs exceeding $1 million. According to Uptime, one in 10 impacted organizations said their last outage had serious or severe impacts. Power issues are still the leading cause of outages, but the risks are evolving: “Failures involving UPS systems, transfer switches and generators are dominant; however, worsening grid constraints and high-density workloads are introducing new pressure points,” according to Uptime. Looking specifically at AI data centers, Lawrence explained that the scale and design of AI-optimized facilities had the potential to increase outage rates in the future. “AI is obviously pushing up rack densities. It’s increasing the demands on the cooling systems,” he said. “It’s reducing the ride through times of cooling. So, we might be entering a period where this line might flatten out or even tick upwards a little bit until we gradually get used to some of these new technologies.” AI Data Centers and Future Resiliency Challenges The Uptime outage report states that, as many AI sites are still relatively new, they are not directly covered by the report but will feature in future research. The report highlights several key questions that need to be addressed: How resilient will large operators choose to make their training sites, given the high cost of infrastructure at scale? The incidence and causes of failures at these sites are not yet well understood. It is possible that AI sites may experience more outages due to the scale and energy intensity of their IT workloads. What systemic risks do very large – or very numerous – data centers pose to regional power grids, and how are operators mitigating those risks? Although mitigation measures are being introduced, the likelihood of systemic failure appears to have increased in recent years and may continue to do so. How resilient will large on-site continuous power facilities prove to be? The growing use of demand response and power trading during periods of stress or peak demand may increase risks. Alex Cordivil, research director at Dell’Oro Group, agreed that the scale of AI infrastructure growth was forcing operators to reconsider their resiliency approaches, but said the process had already begun during the cloud era. “Scale is forcing a rethink of redundancy,” Cordivil told Data Center Knowledge. “We have seen for years how, as scale goes up, operators start rethinking their resilience posture. Hyperscalers have been playing with designs that forego the UPS – arguably the symbol of power resilience in a data center – for the better part of a decade.” The specific technologies used in AI facilities, from liquid cooling to higher-voltage power distribution, also have the potential to impact resiliency due to intrinsic limitations and because they are relatively new – a factor that introduces risk, Cordivil added. “Next up is the broader adoption of direct-current architectures in the power train, which brings a considerable shift in the equipment mix and the introduction of devices that have only hit the market in the last year or two.” Behind-the-Meter Outage Risks Uptime also highlighted the growth in on-site power generation – largely driven by AI – as a potential resiliency risk, which at first could appear counterintuitive given the greater energy autonomy it provides. “On-site power plants are probably going to bring in a whole new class of issues that haven’t up to now existed in data centers,” said Lawrence. “So, if you start to use gas turbines or large amounts of batteries, or fuel cells… they all have their particular issues.” Vlad Galabov, senior research director of enterprise infrastructure for analyst group Omdia,* told Data Center Knowledge that on-site generation infrastructure must be deployed effectively to address some of the resiliency risks inherent to AI. “I can see how having your own source of power can be considered resilient, but it is important to design the whole system around the gas gensets or turbines properly,” he said. For a data center that uses behind-the-meter power to be resilient by design, there are some important considerations related to the spiky nature of AI workloads, explained Galabov. Without these approaches, the generator could be damaged by the unpredictable nature of the AI load. “A pool of capacitors is needed to absorb rapid transient fluctuations which occur all the time during steady-state operation of the GPUs,” he said. Uptime confirmed that it will release more specific research on AI data center resiliency in the near future. An executive summary and webcast on the outage report are available on the company’s website. *Omdia and Data Center Knowledge are both part of Informa TechTarget.