Big Tech plans $750B AI spend

The AI story right now is not really about chatbots. It is about concrete, steel, transformers, cooling systems, and a truly absurd number of GPUs. Over the last two days, Microsoft, Alphabet, Meta, and Amazon updated investors and effectively confirmed that 2026 is turning into a record year for AI infrastructure spending. Add their plans together and you get something in the rough range of $705 billion to $725 billion — with most of that aimed at data centers, servers, networking gear, and the power needed to run them. (cnbc.com) ### Why did this number jump now? Because the companies just refreshed guidance. Microsoft said it expects roughly $190 billion in 2026 capital expenditures, including about $25 billion from higher component pricing. Alphabet raised full-year CapEx guidance to $180 billion to $190 billion. Meta lifted its range to $125 billion to $145 billion. Amazon did not issue the same kind(cnbc.com)llion as AWS demand stays strong. (microsoft.com) ### What are they actually buying? Mostly the boring stuff that turns out not to be boring at all. These budgets are going into AI servers, Nvidia-class accelerators, CPUs, memory, networking, land, buildings, and the electrical gear that lets a data center draw and distribute huge amounts of power. Meta spelled out that part of its increase reflects higher component prici(microsoft.com) — memory and other components got more expensive. (investor.atmeta.com) ### Why does power matter so much? Because the bottleneck has moved. A year ago, the simple story was “who can get enough chips?” Now the harder question is “where do you put them, and how do you feed them electricity?” A modern AI cluster is less like buying more laptops and more like dropping a small industrial plant in(investor.atmeta.com)ied to electrical equipment and data-center plumbing, not just model makers. (forbes.com) ### Are these bets tied to real demand? Yes — and that is the key reason investors are tolerating the spending. Google Cloud grew 63% in the March quarter. AWS grew 28%. Microsoft’s cloud business also kept expanding fast enough that management framed the spending as capacity chasing demand, not speculative overbuilding. In other words, the hyperscalers are not just building for vibes. Enterprise customers are already renting the compute. (cnbc.com) ### Why are investors still nervous? Because even if demand is real, the bill is enormous. Free cash flow gets squeezed when capital spending runs this hot, and pricing pressure on chips and memory makes the math worse. Meta’s stock reaction showed the tension clearly — investors like AI growth, but they do not love open-ended infrastructure bills. The market is basically asking one q(cnbc.com)s? (cnbc.com) ### What does this mean for the rest of AI? It means the center of gravity has shifted down the stack. Consumer AI apps still matter, but the money is concentrating in infrastructure and applied systems — cloud platforms, enterprise tooling, semiconductors, power equipment, and data-center operators. If you want to understand who captures value in this phase, start with whoever controls (cnbc.com) The infrastructure layer gets the checks. (bloomberg.com) ### So what is the bottom line? The headline number sounds like an AI boom story. It is really a capacity race. Big Tech is spending at a scale that makes the old cloud buildout look small — and the constraint is no longer just smarter models. It is whether the industry can build enough powered, cooled, connected physical infrastructure fast enough to keep the whole thing running. (bloomberg.com)