JLL’s newly released 2026 Global Data Center Market Outlook puts hard numbers on what the industry has been calling an “AI infrastructure supercycle,” and the figures are staggering. The report projects nearly 100 gigawatts of new data center capacity coming online between 2026 and 2030, effectively doubling global capacity over a five-year window. The sector is expected to grow at a 14 percent compound annual rate through 2030, with total infrastructure investment requirements approaching 3 trillion dollars and roughly 1.2 trillion dollars in real estate asset value creation.
The economics are shifting just as fast as the volumes. Average global data center construction costs rose from 7.7 million dollars per megawatt in 2020 to 10.7 million in 2025, and JLL forecasts another 6 percent increase to 11.3 million per megawatt in 2026. That figure only covers shell and core. Tenants typically pay for the tech fit-out separately, which can run as high as 25 million dollars per megawatt for AI-grade infrastructure. Rack densities tell the same story. Where racks once topped out at 30 to 40 kilowatts, they are now measured in the hundreds of kilowatts, with some designs approaching the megawatt range, forcing operators to rethink power delivery, cooling, and even basic site selection logic.
The strategic story behind the numbers is a workload transition. AI accounted for roughly a quarter of data center workloads in 2025, with training driving most of that demand. JLL expects AI to represent half of all workloads by 2030, and crucially, expects inference to overtake training as the dominant AI requirement starting in 2027. That shift changes everything: inference workloads are user-facing, latency-sensitive, and geographically distributed in ways training is not. For operators, it means designing facilities for two fundamentally different patterns at once. For nations, it means proximity to users, abundant power, and policy stability are now as critical as raw compute. The result is a global build-out where speed to power, not chip availability, has become the defining competitive constraint.
