Summary
- More data centres are reporting peak rack densities of at least 30kW as accelerated computing changes facility requirements.
- More than half of surveyed organisations struggle to recruit qualified staff, while serious outages remain expensive.
- Water measurement is becoming more common, although carbon tracking and confidence in autonomous control remain limited.
More data centres are reaching peak rack densities of at least 30kW while staffing shortages, power availability, project costs, and outage consequences continue to constrain operations, according to Uptime Institute’s 2026 global survey.
The 16th annual study draws on responses from more than 800 data centre owners and operators. It describes an industry adding capacity while remaining cautious about autonomous control, struggling to recruit experienced staff, and making uneven progress on environmental measurement.
High-density deployments are no longer confined to a small group of experimental AI halls. A growing share of respondents now operates facilities where the highest-density racks reach or exceed 30kW, changing the requirements placed on power distribution, cooling, commissioning, and maintenance.
Peak density changes the wider operating envelope
A facility’s highest rack density does not represent its average load, and many sites now contain conventional air-cooled equipment beside smaller high-density zones. Those mixed environments can be difficult to manage because electrical and cooling systems must support several operating conditions at once.
At 30kW and above, room-level airflow becomes progressively harder to use efficiently. Direct-to-chip liquid cooling, rear-door heat exchangers, contained pods, or hybrid systems may be required, depending on server design and the share of heat that remains in the air.
The transition extends beyond heat removal. Power-distribution units, busways, rack feeds, protection settings, cable routes, and structural loading may all need review, while liquid cooling introduces pumps, heat exchangers, coolant distribution units, water chemistry, leak detection, and a new interface between facility and IT teams.
Retrofits carry additional risk because new pipework and distribution equipment may have to pass through live technical areas. Shutdown windows are limited, customer hardware cannot always be relocated, and commissioning must prove performance without putting existing services at risk.
Power availability and capacity forecasting also rank among the principal concerns identified by the survey. Operators must estimate not only how much aggregate demand will arrive but how quickly customers will occupy high-density halls and whether utility connections, generators, UPS systems, and cooling plant can grow at the same pace.
Installing too much plant early ties up capital and can reduce efficiency at low utilisation, whereas an undersized design can leave a facility with floor space that cannot support the power and thermal requirements of incoming workloads.
Resilience still depends on experienced people
More than half of respondents report difficulty finding qualified candidates, while staff retention remains a persistent problem. The industry is adding electrical, mechanical, and controls infrastructure faster than it is developing people able to commission, operate, and maintain it.
High-density facilities broaden the technical scope of site roles. Teams increasingly need knowledge of electrical protection, liquid systems, water treatment, controls, networking, software, and the boundary between customer cooling equipment and building plant.
Automation can remove repetitive work, but respondents expressed greater confidence in limited uses of artificial intelligence, such as sensor analysis and predictive maintenance, than in autonomous facility control. An incorrect control action in a critical environment can escalate rapidly into a loss of redundancy or an interruption.
Predictive systems also depend on complete and reliable operating data. Sensors require calibration, equipment tags must be consistent, and maintenance records need enough detail to support useful analysis. Poor data can produce false alarms, missed faults, or recommendations that cannot be trusted.
Outage frequency has improved, yet one in ten reported incidents was still classified as serious or severe. The financial and operational cost of those events continues to rise as more critical services are concentrated within individual facilities and customer service-level commitments become more demanding.
Component failure is rarely the entire cause. Procedures, maintenance decisions, change control, monitoring, training, and management oversight determine whether a fault remains contained, which means staffing pressure can weaken resilience even when the physical infrastructure carries sufficient redundancy.
Environmental measurement remains uneven. More than half of respondents now track water consumption, reflecting regulatory and planning pressure around cooling, while carbon-emissions measurement remains less common.
PUE can improve while total energy consumption rises rapidly, and annual WUE can conceal seasonal peaks during periods of heat or water scarcity. Operational reporting will therefore need to combine ratios with absolute resource use and a clearer account of how facilities perform under changing weather and load.
Denser racks are increasing the amount and complexity of plant inside the facility, but the operating model must expand with it. Trained people, reliable data, disciplined procedures, spare parts, and commissioning capacity remain as critical as the additional pumps, busways, and cooling loops.
