Generative AI budgets hit trillions

Generative AI has moved out of the demo phase and into the budget phase. That sounds boring, but it’s actually the real story. The flashy part was chatbots and image models. The expensive part is what came next — giant data centers, power contracts, networking gear, chips, and software teams that can turn those models into something companies will actually use. And in the last week, that spending picture got a lot clearer. ### Why are people suddenly talking in trillions? Because two different trillion-dollar numbers are getting mashed together. One is the upside. McKinsey’s long-cited estimate says generative AI could add $2.6 trillion to $4.4 trillion in annual value across dozens of business use cases — things like customer service, coding, marketing, and R&D. The other is the cost side — what companies need to spend to build enough infrastructure to chase that upside. (mckinsey.com) ### What changed this week? The hyperscalers made the spending real. Bloomberg’s latest roundup put planned 2026 capital expenditures by the biggest U.S. tech companies at as much as $725 billion, driven mostly by AI infrastructure. That jump came after Alphabet and Meta raised their guidance, while Microsoft gave a fresh calendar-year spending view and Amazon held to an already huge plan. (bloomberg.com) ### Which companies are doing the spending? Alphabet now expects roughly $180 billion to $190 billion of 2026 capex. Microsoft’s implied calendar-year figure is about $190 billion. Meta lifted its range to $125 billion to $145 billion. Amazon is still sitting around $200 billion. Add those toget(bloomberg.com) leaning in at once. (cnbc.com) ### Why does capex matter more than model headlines? Because capex tells you what management teams think demand will look like two or three years from now. A chatbot launch can be marketing. A $190 billion infrastructure plan is not. These companies are effectively saying they expect AI demand to stay high enough to justify locking in servers, (cnbc.com)nt a few months ago: consensus 2026 hyperscaler capex had already climbed to $527 billion, and a late-1990s-style peak would imply something like $700 billion. (goldmansachs.com) ### So is the spending actually earning money yet? In some cases, yes. Microsoft said its AI business has already surpassed a $37 billion annual revenue run rate, up 123% year over year. Alphabet’s cloud business is growing fast enough that backlog nearly doubled sequentially to about $462 billion. Th(goldmansachs.com)ng. (microsoft.com) ### What’s the real bottleneck now? Turns out it’s less “can the models work?” and more “can the physical world keep up?” Data centers take years to build. Utilities need to deliver power. Chip supply still matters. The Financial Times captured the mood well months ago — the fear shifted from a compute glut to a power crunch. (microsoft.com)(ig.ft.com) ### Does this mean every company will spend like Big Tech? No — but a lot of enterprise budgets will get pulled toward AI anyway. Most firms won’t build frontier-model infrastructure themselves. They’ll buy access from the cloud giants and then spend internally on copilots, workflow software, data cleanup, security, and engineering changes. Basically, the trillion-dollar headline sta(ig.ft.com)hrough ordinary corporate budgets as companies rework operations around AI tools. (mckinsey.com) ### Bottom line The important shift is simple: generative AI is no longer mainly a story about what the models can do. It’s a story about who is willing to fund the physical and organizational rebuild needed to use them at scale. Right now, the answer is: the biggest tech companies in the world — and they’re writing checks big enough to reshape the rest of the market.