Global data centers hit 110GW
- JLL said global data-center capacity stood near 103 GW in early 2026 and could reach 200 GW by 2030 as AI buildouts accelerate. - The sharpest constraint is power: JLL sees roughly 97 GW of new capacity by 2030, while operators chase grid access, land, and equipment. - The story is less “more servers” than “more electricity” — AI demand is turning power procurement into the core bottleneck.
Data centers are turning into power projects. That’s the real shift. The flashy part is AI — giant GPU clusters, hyperscaler campuses, sovereign compute plans. But the thing actually deciding what gets built now is electricity: who can secure it, how fast, and at what density. JLL’s 2026 outlook put a hard number on that buildout — global data-center capacity at about 103 GW today, heading toward 200 GW by 2030. (jll.com) ### What changed? The new piece of news is the scale. JLL’s January 6, 2026 outlook says the sector could add about 97 GW between 2025 and 2030, effectively doubling global capacity in five years. That is not a normal real-estate cycle. It is an infrastructure surge being pulled forward by AI training, AI inference, and continued cloud expansion all at once. (jll.com) ### Why are people talking about gigawatts? Because megawatts used to be a facility detail. Now they are the product. A data center is still a building full of servers, but for AI the defining question is how much electrical load that building can support. Higher rack densities, liquid cooling, and GPU-heavy deployments mean ope(jll.com) as they shop for chips. Bain says mega-campuses at 1 GW or more are becoming standard for frontier-model training. (bain.com) ### Is AI really the main driver? Basically, yes. JLL expects AI to represent half of all data-center workloads by 2030. ABI Research goes even further on the shape of the shift — it sees AI workloads overtaking legacy workloads by 2031, with more than half of capacity becoming AI-dedicated by then. That m(bain.com)ng of weak sites. (jll.com) ### Why is power the bottleneck? Because demand is rising faster than grids, permits, and utility equipment can keep up. In the U.S., Berkeley Lab’s recent update says data centers used 176 TWh in 2023 and could reach 325 to 580 TWh by 2028. S&P Global’s 451 Research says utility power serving hyperscale, leased, and crypto-mini(jll.com) already has interconnection queues, transformer shortages, and local political fights over land and transmission. (energy.gov) ### So is this a global boom or a U.S. story? Both — but the U.S. still anchors it. JLL says the Americas hold about half of global capacity today. At the same time, the buildout is spreading because power-scarce core markets cannot absorb everything. ABI highlights geographic expansion b(energy.gov)rs in Europe. The map is widening because the old map is crowded. (jll.com) ### What does that mean for operators? The winners may not be the companies with the boldest AI pitch. They may be the ones that can actually execute the boring parts — land control, utility negotiation, cooling design, and backup power strategy. McKinsey and Bain both frame the next phase as less of a blind scramble and more of(jll.com), but site readiness matters more than it used to. (mckinsey.com) ### What’s the catch? Forecasts are all over the place. The numbers differ because analysts measure different things — IT load, utility demand, leased capacity, global versus U.S. markets. That is why a viral “110 GW” figure can sound contradi(mckinsey.com)ry model. (msn.com) ### Bottom line The data-center boom is no longer just a tech story. It is an electricity story with concrete, copper, transformers, and permits at the center. AI is pushing the industry toward much larger campuses and much higher densities — but the market is starting to learn that compute demand is the easy part. Getting the power is the hard part.
