Uptime Survey Shows AI Capacity Planning Risk For Data Centres
Uptime Institute survey data in Data Center Knowledge shows AI pushing operators toward phased capacity decisions as costs, forecasting, power availability and skills pressure converge.
AI data centre growth is forcing operators to plan for uncertainty, not a single expansion curve.
Uptime Institute's 2026 survey, covered by Data Center Knowledge, placed cost control, capacity forecasting, power planning and staffing in the same operator-risk frame, even where operators are not yet running large training clusters.
The operating issue is not only whether more facilities will be built.
Operators have to decide how much power and cooling to install before they know whether a site will support high-density AI clusters, conventional enterprise workloads or a mix that changes after construction starts.
Costs And Forecasting Move Together
Uptime's survey was collected in April and May 2026 from a broad owner and operator sample.
According to Uptime's survey results, cost remained the leading management concern while capacity forecasting moved close enough to shape the same approval process.
The Uptime survey figures put cost concern at 79% of operators, capacity-forecasting concern at 76%, supply-chain disruption concern at 72% and power-availability concern at 64%.
Those constraints now sit in the same management discussion instead of in separate engineering, procurement and finance tracks.
That grouping changes how new capacity is approved.
A site plan that solves space without solving equipment availability, utility service and staffing exposure can leave a developer with expensive capacity that cannot be brought online at the assumed pace.
Rack Density Splits The Market
The density signal points to uneven growth across the installed base.
The Uptime survey account put modal rack density at 11 kW, up from 9 kW in 2025 for the most common configuration.
Those numbers point to a split between AI training sites and the broader enterprise estate.
Uptime Institute chief technical officer Chris Brown described AI factories as the part of the market lifting density fastest, while CPU-based enterprise workloads still account for most installed capacity, Data Center Knowledge's account showed.
Phased construction becomes the hedge in that environment.
Operators can open a facility around current demand, then add electrical and cooling capacity as workload commitments become clearer, instead of assuming every new hall should be built immediately for AI-scale density.
Third-Party Sites And Skills Add Pressure
Workload location is shifting at the same time.
The survey figures showed 46% of IT workloads now running in third-party facilities such as colocation and cloud, narrowly ahead of 44% in enterprise-owned sites for the first time.
That handoff does not remove operational risk; it moves it into contracts, service tiers and provider capacity.
Enterprises relying on external facilities still need to know whether their provider can secure power, cooling, outage resilience and expansion room as AI demand competes for the same infrastructure.
The staffing signal makes the planning problem harder.
The Uptime survey account put qualified-candidate difficulty at 53% of operators, compared with 46% in 2025.
A shortage of qualified staff can slow commissioning, maintenance and incident response even when capital and utility capacity are available.
Fewer Outages Do Not End The Reliability Question
Reliability data offered a partial counterweight.
The Uptime survey figures put the share of operators experiencing an impactful outage in the previous three years at 47%, down three percentage points year over year, while serious or severe outages stayed at 10% of outages.
That improvement does not erase the next risk because AI infrastructure changes the operating envelope.
Higher densities, liquid-cooling adoption, grid instability, extreme weather and skills constraints all reduce the margin for slow decisions when power or cooling problems appear.
Sustainability tracking shows the same pattern of progress and gaps.
The survey figures put power-consumption tracking at 87% of operators and PUE tracking at 79%, while Scope 3 emissions tracking stood at 21%.
AI may be raising the urgency of capacity decisions, but customers and regulators still need visibility into the energy and supply-chain consequences of those decisions.
AI capacity planning cannot follow a straight real-estate expansion curve.
Power, cooling, staffing, provider contracts and measurement systems now determine whether a data centre build can turn AI demand into usable and resilient capacity.
