The electricity finally showed up to the party, and it's about to bounce the guests.

The Summary

The Signal

Data centers have been the invisible backbone of the AI boom. Now they're becoming very visible, and neighbors aren't happy. Local governments are starting to impose restrictions on new construction as residents push back against the power consumption, cooling water usage, and industrial footprint of facilities that can draw as much electricity as a small city.

This matters because the agent economy runs on compute, and compute runs in buildings. Every foundation model training run, every inference cluster serving millions of AI agents, every robotics simulation optimizing warehouse flows needs physical infrastructure. The companies building Web4 have largely assumed that if they had the chips and the capital, the rest would sort itself out.

"Restrictions may slow construction without derailing the broader buildout."

That assumption is breaking. Crescenzi's framing is telling. He's not worried about a full stop, he's modeling for friction. Longer permitting timelines. Higher costs for grid upgrades. More strategic site selection. This is the maturation of an industry that thought it could scale like software but is learning it scales like utilities.

The political economy of data centers is straightforward:

  • Hyperscalers want cheap power and proximity to fiber
  • Communities see tax revenue but also strain on infrastructure built for different loads
  • State governments want the jobs and investment but can't magic up gigawatts

What changes is the timeline and the cost structure. If permitting adds 12-18 months to a build cycle, that's capital sitting idle. If grid connection fees triple because utilities need substation upgrades, that flows straight to the economics of training runs and inference pricing. The companies with existing capacity, with relationships, with sites already permitted, just got more valuable.

The Implication

Watch for a bifurcation in AI infrastructure. The hyperscalers with deep pockets and government relationships will navigate this. Smaller players will either pay premiums for colo space or get creative with edge compute and distributed training. The real winners might be the companies that figured out power efficiency early, not just raw performance. If building new data centers gets harder, making better use of existing ones becomes the edge.

Sources

Bloomberg Tech