Is the AI Buildout a Bubble?

Is the AI buildout the fiber-optic bubble with better marketing — or the next industrial revolution: a once-a-century buildout where the expensive mistake is standing aside? Somewhere between those two convictions sits nearly every allocator, executive, and sideline observer in America, most of them arguing from priors rather than evidence. The bear case writes itself: $710 billion of capital budgets in a single year, spending racing ahead of operating cash flow, circular deals between suppliers and customers, and an accounting fight over whether the chips hold their value — all of it rhyming with 2000. The bull case is just as loud: demand the builders cannot construct fast enough to meet, the most profitable companies in history writing the checks, and chief executives insisting the greater risk is building too little, too late. They cannot both be right. And in May and June 2026, the market quietly began producing what this debate has always lacked: hard, checkable evidence — signed contracts with disclosed prices, a rental market that reprices aging chips daily, and disclosures that cut against the loudest claims on both sides. Which way the evidence points, and how decisively, is the subject of this brief: whether the buildout clears its cost of capital, the honest case against, and the precise scorecard for the July 22–30 earnings window. If you own the trade, are avoiding it, or cannot decide, the next 4,000-plus words settle what can be settled — and name exactly what to watch for the rest.