Big Tech spends $725bn on AI

Data centers are the story here. Not chatbots, not demos, not glossy product launches — concrete, power, networking gear, and a mountain of GPUs. In late-April earnings, Alphabet, Amazon, Meta, and Microsoft made clear that the AI race has turned into a buildout race, with combined 2026 capital spending now landing around $725 billion. That number matters because it shows the industry has moved past experimentation. The big platforms are now rebuilding the physical internet around AI. ### Why is the number so big? Because AI at hyperscaler scale is brutally physical. Training frontier models needs giant clusters of accelerators. Serving those models to millions of users needs even more chips, memory, networking, and cooling. Then you need the buildings, substations, fiber, and backup power to keep the whole thing running. This is less like launching an app and more like laying railroad track — except the track goes obsolete fast, so you keep relaying it. (bloomberg.com) ### Who is spending what? Microsoft gave the cleanest single figure — roughly $190 billion in calendar 2026 capex, including about $25 billion from higher component pricing. Alphabet lifted its full-year 2026 capex guide to $180 billion to $190 billion. Bloomberg’s rollup put Amazon at about $200 billion and said it was the only one of the big four not to raise its figure after earnings. Meta’s first quarter alone included $19.84 billion of capex, which tells you how fast its infrastructure machine is running. (aws.amazon.com) ### Why did investors flinch? Because revenue is growing, but cash is getting eaten first. Amazon’s March-quarter spending already cut into free cash flow. Meta generated $32.23 billion of operating cash flow in Q1 2026, but free cash flow was only $12.39 billion after capex. That is the basic tension — AI can be lucrative, but before it throws off cash at scale, it soaks up cash at scale. (microsoft.com) ### Why do buybacks matter here? Buybacks were one of the quiet engines under the bull market. Goldman’s read, echoed widely after this earnings round, is that AI capex is now crowding them out. S&P 500 capex growth is tracking at 39% in Q1 2026, versus just 1% for gross buybacks. Goldman also expects S&P 500 capex to rise 33% in 2026 while buybacks grow only 3%. For hyperscalers, the squeeze is sharper because their spending is so concentrated. (bloomberg.com) ### Is this just a bubble signal? Maybe partly — but not only that. The stronger case is that the cloud leaders think AI demand is real enough that underbuilding is riskier than overbuilding. If Microsoft or Amazon cannot supply compute, customers go elsewhere. If Alphabet falls behind on AI infrastructure, it risks both cloud share and product momentum. The companies are spending like landlords in a land rush because the first rule is simple: don’t run out of capacity. (finance.yahoo.com) ### What gets built with all this money? Mostly the boring stuff that ends up deciding who wins — data centers, custom silicon, networking, storage, and power systems. AWS is openly framing its infrastructure as built for AI workloads, while Microsoft and Alphabet keep tying spending to cloud and AI demand. Meta is doing the same from the consumer side, where recommendation systems, ads, assistants, and model training all feed the same infrastructure flywheel. (microsoft.com) ### So what is the real takeaway? The AI race is no longer mainly about better models. It is about who can finance, build, and operate the biggest compute estate without breaking their cash machine. $725 billion is the headline, but the deeper point is that Big Tech has decided AI is important enough to trade away some buybacks, some free cash flow, and a lot of short-term comfort to get there first. (bloomberg.com) (aws.amazon.com)