Power delays stall data-center builds

Data centers are running into a very old-economy problem. Not chips. Not servers. Not even money. Power. That is the real story behind the latest wave of AI buildouts in the U.S. Demand is still ripping. Leasing is still strong. But the thing slowing projects down is the wait for enough electricity to show up at the site — plus the substations, transformers, transmission upgrades, and utility approvals needed to make that power real. In 2026, that has become the main bottleneck. (cbre.com) ### Why is power the choke point? A modern AI campus is not a normal data center scaled up a bit. It is a huge electrical load. JLL says power — not location or even cost — is now the primary site-selection criterion because grid connections come with multiyear waits. CBRE makes the same point more bluntly: getting 300 MW-plus delivered in under 36 months now matters more than pure connectivity. (jll.com) ### What changed so fast? AI changed the shape of demand. The market is no longer centered on smaller facilities that could be delivered on old 12-to-18-month schedules. CBRE says those timelines no longer hold, especially as projects shift toward 500 MW-plus (jll.com)ush interconnection timelines to 24, 36, or even 48+ months. (cbre.com) ### Is demand really that strong? Yes — and that is why the delays matter so much. CBRE says vacancy in North America’s primary markets fell to a record-low 1.4% at the end of 2025. Net absorption hit a record 2,497.6 MW in 2025, beating the prior year’s record. JLL says global occupancy has reached 97%, which means there is very little slack left in the system while new projects wait on power. (cbre.com) ### So what does the bottleneck look like on the ground? It means projects that look financeable on paper miss their deployment windows in real life. Data Center Frontier describes large-load interconnection queues as the main reason otherwise viable AI campuses are slipping. Developers are being told capacity w(cbre.com)r alternate energy plans. Even a short delay hurts economics — one example in that piece shows a 60 MW AI facility’s projected return falling sharply after just a few months of delay. (datacenterfrontier.com) ### Why can’t utilities just speed up? Because this is not one switch flip. A big data center can require transformers, switchgear, transmission upgrades, generation planning, and queue studies across regional grid operators. Data Center Front(datacenterfrontier.com)terconnection and transmission processes exist, yet they take years to show up in actual project timelines. (datacenterfrontier.com) ### What are builders doing instead? They are moving earlier and getting more creative. JLL says tenants now need to secure power in parallel with real-estate decisions, not after. CBRE says preleasing is staying unusually high, and developers(datacenterfrontier.com), storage, and private energy arrangements — as a way to bypass the slowest parts of the grid queue. (jll.com) ### Does this change where data centers get built? Basically, yes. Sites without firm power are becoming less valuable no matter how good the land looks. JLL warns that assets without secured power are increasingly unviable. CBRE says emerging markets and deregulated states are gaining appeal, especially where developers can line up generation faster or avoid the worst congestion in legacy hubs. (jll.com) ### Bottom line The constraint on AI infrastructure is no longer mainly capital or demand. It is time-to-power. Until utilities, grid operators, and developers solve that, the U.S. data-center boom will keep growing — but in a slower, more uneven, and more expensive way. (cbre.com)