AI is turning Big Tech into Big Capex. The market still wants to treat AI like a software story, but the spending profile increasingly looks like an industrial buildout. That distinction matters. For most of the last two decades, the best tech businesses had a simple advantage: scale without much incremental capital. Microsoft could sell another copy of software, Google could serve another search, Meta could add another user, and the marginal economics were absurdly good. Even Nvidia, the physical-product outlier, kept the dirtiest capital intensity outside its own walls by outsourcing manufacturing. That was the low-asset edge. Now the hyperscalers are moving in the other direction. Goldman has AI hyperscaler capex at roughly $755 billion this year, essentially absorbing all operating cash flow, while assets have grown 138% in five years. That is not a rounding error. That is a change in the business model investors are underwriting. Gregg Fisher’s line is blunt: “We know from this 100 years of data that capex is bad.” Not because all investment is stupid. Because history says companies that need to keep feeding assets usually deliver worse market returns as a group than companies that can scale without constantly rebuilding the balance sheet. This is the uncomfortable part of the AI boom. The CEOs are not idiots. Satya Nadella, Sundar Pichai, Mark Zuckerberg, Andy Jassy — these are not executives casually torching cash because they forgot capital discipline. They are spending because the perceived cost of underinvesting is existential. If AI creates winner-take-most economics, being second or third may mean getting structurally demoted. So everyone spends like the downside of restraint is bigger than the downside of waste. That is rational at the individual company level. It can still be ugly at the industry level. Bezos calling AI an “industrial bubble” is probably the cleanest framing: society can win, the technology can be real, and investors can still overfund the buildout. Those ideas are not contradictory. The mistake is assuming every dollar of AI capex earns monopoly-like software returns. Some will. A lot will not. Bottom line: AI is not just a technology story anymore. It is a capital-allocation story, and the market is starting to price Big Tech like it can spend like an industrial while still compound like software.