Active Data Center IT Capacity by Workload: 2025 to 2035

Key Takeaways:

Global active data center capacity grows at a 23.8% CAGR from 2025 to 2035, expanding from 24.4 GW to 147.1 GW as AI workloads layer on top of continued cloud growth.

AI-dedicated capacity grows faster than legacy workloads, reaching 94.8 GW by 2035 and becoming the majority of global active data center capacity in the early 2030s.

Legacy workloads remain a significant share of capacity in the near term, with AI overtaking legacy between 2030 and 2031 due to slower enterprise AI migration.

Tier One hyperscalers and neoclouds leave increasing headroom between active and available capacity, reflecting the need for new AI use cases to fully utilize deployed infrastructure.

Power availability, grid access, and permitting emerge as the primary constraints on data center expansion by the early 2030s, replacing server demand as the main limiting factor.

About This Research

This is a first-of-its-kind forecast explicitly modeling data center capacity rather than Information Technology (IT) spend or server shipments. By translating power envelopes into active, available, and under-the-roof capacity, the forecast addresses a structural gap in existing industry datasets, which typically stop at Capital Expenditure (CAPEX), Megawatt (MW) announcements, or hardware volumes.

The model captures capacity states that are not visible in a conventional market forecast. Distinguishing between deployed (active), reserved (available), and under-the-roof capacity and further separating AI-dedicated workloads from legacy workloads, provides a clearer view of AI readiness, infrastructure slack, and utilization dynamics that traditional utilization, server, or revenue-based metrics cannot reveal.

This forecast responds to the growing disconnect between AI infrastructure investment and observable workload deployment. With AI-driven datacenter announcements accelerating faster than realized demand, forecasting capacity (rather than just spend) provides a more grounded way to assess whether infrastructure build-out is running ahead of viable AI use cases.

This forecast illustrates the changing bottlenecks in data center expansion. Power availability, grid connection timing, and permitting constraints now shape capacity trajectories more than server demand alone, making a MW-based forecasting approach increasingly necessary.

The forecast enables forward-looking analysis that is not currently available elsewhere in the industry. Linking capacity growth to operator type, workload mix, and regional power dynamics allows stakeholders to assess future constraint risk, utilization ceilings, and AI scalability beyond what existing market data products offer