The global data center sector is on track to double in size by 2030, and the reason isn't training more models. It's serving them. JLL's 2026 Global Data Center Market Outlook puts new capacity at roughly 100 GW between 2026 and 2030, lifting installed capacity toward 200 GW, and points to 2027 as the year inference workloads overtake training as the dominant AI demand.
That timing matters for anyone planning cloud or edge capacity. Training runs in a handful of giant clusters. Inference has to run close to users, which pushes compute out of central campuses and into regional hubs and edge sites. The geography of cloud infrastructure is about to change, and not in the way most operators expected.
The stakes are large enough that this is no longer a niche facilities question. JLL estimates total data center spending approaches $3 trillion by 2030, including $1 to $2 trillion tenants will spend on GPUs and networking gear on top of $1.2 trillion in real estate value creation. Energy has become the binding constraint, with grid connection wait times in primary markets exceeding four years.
Inference flips the workload map
AI accounted for about a quarter of all data center workloads in 2025, and training drove most of that demand. JLL expects AI to reach half of all workloads by 2030, with inference becoming the primary driver around 2027. The shift follows a simple economic logic: training a model is a one-time or periodic investment, while inference generates ongoing revenue as users actually use the model.
Every deployment creates sustained inference demand that grows with adoption. That demand, however, depends on inference applications that don't yet exist at scale. If those applications arrive slowly, the capacity math changes.
Inference also needs geographical distribution to cut latency and serve users effectively. That pushes workloads from centralized training clusters into regional hubs and embedded systems at the edge. Cloud providers that spent a decade consolidating into a few mega-regions now face pressure to put capacity closer to users.
Power, not land, is the bottleneck
The average wait for a grid connection in primary data center markets now exceeds four years. Operators are responding with behind-the-meter power arrangements and colocated battery storage rather than waiting in the utility queue.
Natural gas is projected to play a major role in the U.S., both for temporary bridge power and increasingly for permanent on-site generation, evidenced by surging global turbine orders. Some of the largest tenants reject natural gas as unsustainable, so the approach won't be universal. In EMEA and APAC, renewables such as solar and wind are seeing increased use. In EMEA, projects combining renewables with private wire transmission can cut tenant power costs by 40% compared to the grid.
Several markets have moved to "bring your own power" mandates, including Ireland and Texas. Operators are also moving beyond power purchase agreements to fund their own generation directly. Site selection now ranks speed to power first, ahead of community support, latency, and proximity to customers.
Construction costs keep climbing
Between 2020 and 2025, average global data center construction cost rose from $7.7 million to $10.7 million per MW, a 7% compound annual growth rate. JLL forecasts $11.3 million per MW for 2026, a 6% increase. Those figures cover shell and core only. Tenants can spend up to $25 million per MW on AI tech fit-out, which dwarfs the building itself.
The relentless pace of expansion has produced extended lead times and limited skilled trades. Projects are getting larger and more expensive, and the sector continues to consolidate as development costs and operational sophistication rise. Barriers to entry are removing speculative projects from the development queue and accelerating viable projects backed by credible companies.
For those groups, debt markets remain open. JLL estimates roughly $870 billion in new debt financing is needed to fund the new supply.
Regional math: Americas leads, APAC grows fastest in share
The Americas is the largest region at about 50% of global capacity, with the U.S. accounting for roughly 90% of capacity in the Americas. The region is projected to grow at 17% supply CAGR through 2030, preserving its dominance.
APAC capacity will expand from 32 GW to 57 GW by 2030 at a 12% CAGR. Colocation leads growth at 19%, while on-premises capacity is projected to decline 6% as enterprises continue migrating to cloud. EMEA's 10% CAGR is fueled by government support for AI infrastructure and strong demand for sovereign AI clouds to meet data privacy regulations, with 13 GW of new supply concentrated in established European hubs and emerging Middle Eastern markets.
The on-premises decline in APAC is worth watching. Enterprises in the region are shifting workloads to colocation and cloud providers rather than building their own facilities, which concentrates demand among fewer, larger operators.
What this means for cloud and edge buyers
The transition from training to inference redistributes workloads from centralized clusters to distributed regional hubs. Capacity planning and geographic deployment strategies need to account for that now, not after latency problems appear in production.
For edge specifically, the implication is direct. If inference is half of workloads by 2030 and it needs to run near users, edge sites and regional points of presence stop being optional. Cloud providers will need to place capacity in markets they previously served from a distance. The winners of this cycle will be those who can anticipate demand inflection points while maintaining flexibility as AI models and use cases evolve.
It's not clear that all the announced capacity will actually get built. Grid constraints, rising construction costs, and tenant reluctance on natural gas all point to a gap between planned and delivered capacity. Buyers should plan for tighter availability in 2027 and beyond, and should not assume pricing stays flat.
For more on how data center real estate is shaping cloud strategy, see the JLL outlook at JLL and the related coverage in our cloud and edge section.
One more thing to watch: construction cost inflation. At 6% in 2026 on top of 7% annually since 2020, every megawatt of new capacity costs more than the last. That compounds across a 100 GW buildout and shows up in lease rates eventually.