Twenty-three gigawatts of data center capacity is currently under construction worldwide. That’s not a projection. It’s what was already in progress as of September 2025, per BloombergNEF’s analysis released March 24, 2026. For context: 23 GW is roughly equivalent to the entire electricity generation capacity of a mid-sized European country. It’s all being built for AI workloads.
The capital behind that build-out is substantial. BloombergNEF estimates the combined capital expenditure of the 14 largest publicly owned data center operators could reach approximately $750 billion in 2026. Analyst estimates vary, Futurum Research puts a related figure at $690 billion, measuring a different scope. Both are analyst projections, not confirmed spending totals. The directional signal is consistent across both: this is a capital commitment at a scale that hasn’t been seen in infrastructure investment outside of national energy grid buildouts.
The Americas region is carrying the largest share of the build-out. BloombergNEF data places 17 of the 23-plus GW across 311 locations in the Americas. That concentration isn’t coincidental, it reflects the clustering of major hyperscaler headquarters, existing grid infrastructure, and available land in the United States.
Energy is where this story gets complicated. The IEA projected in April 2025 that global AI data center energy consumption would grow from 485 TWh in 2024 to 945 TWh by 2030, a 95% increase in six years. That projection, cited by Climate Home News, places AI data centers at approximately 3% of global electricity demand by 2030. To be clear: this is an IEA projection from data that’s now a year old. Projections in a fast-moving market carry uncertainty. But the IEA’s track record on energy demand forecasting gives this estimate credibility beyond what a single analyst report would carry.
The grid pressure point is real. Texas’s ERCOT system has been navigating extraordinary data center power requests. The specific figures from that reporting haven’t been independently confirmed in this cycle and are excluded here, but the grid capacity question is documented in multiple independent sources and represents a genuine constraint on the build-out’s pace.
What to watch: Three forward indicators are worth tracking. First, utility capacity agreements in major data center markets, when hyperscalers sign power purchase agreements at scale, it signals committed rather than planned spending. Second, any regulatory action on data center energy consumption in the EU, where grid stress is politically visible. Third, the relationship between construction starts and capacity delays, 23 GW under construction doesn’t mean 23 GW delivered on schedule. Power permitting timelines are the friction point.
TJS Synthesis: This is infrastructure investment at a scale that reshapes energy markets, not just technology markets. The $750 billion capex estimate and the IEA’s 95% energy growth projection tell the same story from different angles: AI development has a physical footprint that its proponents rarely foreground and its critics rarely quantify well. Both the capital concentration and the energy demand trajectory are now large enough to attract regulatory attention. When that happens, and it’s a when, not an if, the data center build-out becomes a policy story as much as a technology story. This brief pairs with the Meta data center analysis published today and the broader AI infrastructure deep-dive linked below.