Off the Run

The Grid Can’t Wait and Neither Can America

• The $700 billion AI capex cycle is real. The opportunities for returns are not in question.
• Power is the governor. Not chips, not demand, not balance sheets — electrons.
• America’s permitting system was designed to stop things from being built. Private capital is finding ways to build.

The hyperscalers and their $700 billion capex supercycle are sucking all the air out of the room. The AI and data center trade is driving the market to all-time highs, and with a war still grinding in Iran and a US-China summit looming, plenty of people are asking whether any of this is justified or just waiting for the bottom to fall out. There is no denying the capex cycle is real. These companies have the cash, they have the demand, and the returns are starting to show up. But the market may be asking the wrong question. It is not whether AI spending generates returns. It is how quickly that capital can actually be deployed when the thing it needs most is not on a balance sheet.

So where exactly is the $700 billion going? Chips, networking gear, data center shells, cooling systems, and increasingly, power generation secured outside the grid because the grid cannot move fast enough. Again, the spending is real. The returns, however, require a sequence of physical events that has a well-documented habit of running late. Of the 16 gigawatts of data center capacity slated to come online this year, only about 5 gigawatts is currently under construction. Last year, 26% of expected capacity slipped, and another 10% quietly pushed back without much notice. Microsoft alone has disclosed an $80 billion backlog of Azure orders it cannot fulfill because of power constraints. The demand is real, the contracts are signed, the customers are waiting, and the thing standing between Microsoft and $80 billion in revenue is electrons. The hyperscalers are promising investors a roughly 24-month return on this capital. That math assumes the power shows up on time. The track record suggests it often does not.

This is not a Microsoft problem. It is an American one. America is trying to win an AI race with a permitting system designed to stop things from being built. Meanwhile, China spent $500 billion on energy in 2025 alone and added power capacity at roughly eight times the US rate. It is clear that both parties want permitting reform. And yet, the interconnection queue sits at 2,600 gigawatts, the median wait time is approaching five years, and Google has flagged potential delays of up to 12 years. What America needs is an Energy Industrial Policy. If that sounds too much like central planning, call it strategic coordination. Either way, the job is the same: align generation, transmission, storage, supply chains, and permitting around one national objective. We have not had anything like that in a generation because for decades the answer was to send the hard stuff somewhere else. Now we want it back on a two-year political cycle. That is not a plan.

The good news is that progress does not require waiting for Washington. Behind the meter generation is growing fast, with data centers increasingly pairing directly with power sources rather than waiting years for grid connection. Long duration battery storage is starting to make that model more reliable and more bankable. Pine Island, where Google, Xcel Energy, and Form Energy are working together to firm up renewable power for data centers using iron-air batteries, is one example of what coordinated private investment can accomplish without a policy catalyst. Private capital is stepping into gaps that utilities and regulators cannot fill quickly enough. And in that environment, owning a utility starts to look very different than it did five years ago. Not as a regulated, slow moving yield vehicle. As a platform that already has what everyone else is waiting years to get: the interconnection, the customers, the regulatory standing, and the ability to say yes when a hyperscaler comes knocking.

What’s past is prologue. Every time America faced a generational infrastructure challenge it eventually found a way to treat it as a national imperative. The highway system. ARPANET. The space race. In each case the strategic stakes were clear enough that the political noise eventually gave way. The AI buildout may be the same test. The capital is ready. The demand is real. The technology exists. The only thing missing is the thing you can’t build on a balance sheet.