Corporate data centers continue to lose workload share to colocation and cloud providers. Meanwhile, data center operators face mounting challenges tied to AI adoption, including increasingly difficult capacity planning, persistent staffing shortages, and more expensive outages.
Third-party data center facilities and services—including colocation, cloud, hosting, and SaaS providers—account for 46% of IT workloads in 2026, edging past the 44% in enterprise-owned corporate data centers, according to Uptime Institute’s 16th Annual Global Data Center Survey. This is the first time in Uptime’s annual survey history that third-party data center venues have held the larger share of enterprise IT workloads than corporate-owned facilities. Uptime expects the gap to widen to 48% versus 42% by 2028, although it says enterprise facilities will remain a key part of hybrid IT strategies, with security and regulatory requirements continuing to keep many workloads on premises.
Uptime’s survey, conducted between April and May, collected responses from more than 1,600 participants worldwide, including more than 800 data center owners and operators.
“All of these are linked, and they’re all linked through AI,” Andy Lawrence, Uptime’s executive director of research, said during a webinar discussing the findings. “The pressures it’s putting on the entire supply chain, the cost base, the availability of power—all of these are being driven by this massive surge in demand for AI infrastructure.”
Cost remains management’s top concern for the fourth straight year, with 79% of respondents saying executives are at least somewhat concerned. But the share describing management as “very concerned” fell to 38%, from 44% last year.
The results suggest operators are balancing cost concerns against a broader set of AI-related challenges. Concern about forecasting future capacity requirements climbed to 76%, while supply-chain disruptions, staffing shortages, power availability, and meeting AI infrastructure needs all ranked among management’s leading worries.
“Cost issues kind of fell slightly, but that doesn’t mean that people are no longer concerned,” Lawrence said. “It just means the other issues got serious.”
As organizations move beyond experimentation with generative AI, operators appear to be taking a more thoughtful view of where the technology can be used safely.
The share of respondents who believe AI improves facility efficiency slipped to 52% from 58% last year, while confidence that AI reduces human error fell 11 percentage points to 40%. Trust remains strongest for lower-risk applications such as operational analytics and predictive maintenance but drops sharply for autonomous operational tasks. Just 31% trust AI to control equipment settings, and only 16% are comfortable with automated configuration changes.
“This drop in trust … suggests that a lot of operators are nowhere near the point of handing over a lot of control tasks to AI,” Jacquie Davis, Uptime research analyst, said. “It speaks to the importance of human in the loop.”
AI deployments are also beginning to reshape infrastructure design.
Average modal rack density surpassed 11 kW for the first time, a milestone driven largely by a growing number of AI-focused facilities with typical rack densities above 30 kW. Excluding those deployments, average rack density increased more modestly, from 7.5 kW to 7.8 kW. Nearly one-quarter of operators now report at least some racks exceeding 30 kW, up from 19% a year ago.
Daniel Bizo, Uptime’s research director, called surpassing 11 kW a “psychological milestone” for the industry because the average now reflects the growing influence of AI-oriented deployments. Chris Brown, Uptime’s chief technical officer, said AI training workloads are driving the highest rack densities, while more traditional enterprise applications remain the primary driver for facilities below the 30 kW range.
“AI is driving rack densities up faster than anywhere else,” Brown said, adding that enterprise applications remain the dominant driver for most facilities operating below the highest-density AI deployments.
For the sixth consecutive year, fewer operators reported experiencing an impactful outage, with 47% saying they had one during the previous three years, down from 50% in 2025. But outages continue carrying a higher financial toll. Seventy-one percent of operators that experienced an outage said their most serious incident cost at least $100,000, up from 57% last year.
Power failures remained the leading cause of impactful outages, accounting for 56% of incidents. Brown said cooling-related failures are likely to become more significant as AI-driven rack densities continue increasing.
After stalling in last year’s survey, environmental reporting resumed its gradual upward trend. More than half of operators now track water consumption, while Lawrence said the temporary slowdown reflected the industry’s intense focus on expanding AI capacity rather than any retreat from sustainability goals.
More than half of operators now track water consumption, up from 47% a year ago, suggesting that even as AI infrastructure dominates investment decisions, environmental reporting remains a growing priority across the industry.
The industry’s talent shortage also intensified after several years of relative stability.
More than half of respondents reported difficulty finding qualified candidates, up from 46% last year, while 28% said they had lost employees to competing data center operators. Electrical specialists and junior operations staff were the most difficult positions to fill, reflecting growing demand for AI-related infrastructure expertise.