Big Tech $725bn AI spend

Big Tech’s AI boom has stopped looking like a software story and started looking like an infrastructure story. That is the real shift hiding inside the big headline numbers. Microsoft, Meta, Alphabet and Amazon are still growing fast, but they are also pouring cash into data centers, chips, networking and power at a pace that would have looked absurd two years ago. The news this week is that the combined spend for this buildout has climbed to roughly $725 billion over the current investment cycle, while free cash flow has been pushed down to levels not seen in about a decade. (ft.com) ### What are they actually buying? Mostly not “AI” in the abstract. They are buying the physical stack that makes AI possible — GPU clusters, custom accelerators, servers, networking gear, cooling systems, backup power, and the buildings to hold all of it. The old hyperscaler playbook was already capital intensive, but generative AI made the density problem much worse. Training frontier models is expensive. Serv(ft.com)ive too. And the catch is that inference — the cost of answering prompts all day — can become a permanent margin drag, not a one-time launch cost. (ig.ft.com) ### Why does the $725 billion number matter? Because it changes the kind of companies these are. Big Tech used to be admired for acting like asset-light cash machines — huge margins, low capital intensity, tons of free cash flow. That is no longer the whole picture. If you have to keep building AI “factories” just to stay in the race, then investors start judging you less like pure software and more li(ig.ft.com)is a very different valuation conversation. (ft.com) ### Who is showing the strain most clearly? Meta and Microsoft make the tension easiest to see. Meta’s first-quarter 2026 revenue jumped 33% to $56.3 billion, but capital expenditures were also $19.84 billion in just one quarter, leaving free cash flow at $12.39 billion. Microsoft’s March-quarter revenue rose 18% to $82.9 billion, and it said its AI business surpassed a $37 billion annual revenue run rate, up 12(ft.com)rs — but Microsoft also said gross margin was pressured by continued AI infrastructure investment and growing AI product usage. (investor.atmeta.com) ### What about Alphabet? Alphabet is still printing cash, but it is leaning harder into the same buildout. On its April 29, 2026 earnings call, the company raised full-year 2026 capex guidance to $180 billion to $190 billion. At the same time, Google Cloud revenue grew 63% and topped $20 billion for the first time, while backlog nearly doubled quarter over quarter to more than $460 billion. Basically — demand is real, but so is the bill. (abc.xyz) ### So is this irrational overspending? Not obviously. The companies keep making the same argument: demand is outrunning supply. If that is true, underbuilding is as dangerous as overbuilding. Lose capacity, and you lose model training speed, enterprise contracts, developer mindshare, and maybe the whole platform layer. But the market is (abc.xyz)ns into durable revenue, not just flashy usage. (ig.ft.com) ### What changes inside these companies? Product decisions get harsher. A feature that drives engagement but burns expensive inference may not survive. Hiring changes too — more weight on infrastructure efficiency, chip utilization, model compression, and pricing discipline. The internal question stops being “can we launch this?” and becomes “can we launch this at a gross margin that makes sense?” That is a very different operating culture. (microsoft.com) ### Why should anyone outside Big Tech care? Because this buildout spills into the rest of the economy. It affects demand for semiconductors, power, real estate, fiber, and construction. It also shapes what AI products cost and which ones stay free. If compute remains scarce and expensive, the internet starts to feel less like software abundance and more like metered infrastructure. (ig.ft.com) ### Bottom line? The AI race is no longer just about smarter models. It is about who can finance, build and monetize the biggest machine behind them. The winners will not be the companies that spend the most. They will be the ones that turn giant capex into profitable, repeatable compute economics.