Reports say Google preparing up to $40B package for Anthropic

This is an AI infrastructure story disguised as a funding round. Google and Anthropic said on April 24 that Google plans to invest up to $40 billion in the startup, with $10 billion going in now and as much as $30 billion more tied to performance milestones. But the bigger thing is the hardware commitment wrapped around it — Anthropic is locking in 5 gigawatts of next-generation Google TPU capacity starting in 2027. (cnbc.com) ### Is the $40 billion report real? Yes — at least the broad outline is. Google and Anthropic publicly confirmed the “up to $40 billion” figure on April 24, 2026. The social posts weren’t inventing the deal from scratch. What they were doing was compressing a more complicated package into a single dramatic numbe(cnbc.com)lestones. (cnbc.com) ### So what is Google actually buying? Part ownership, obviously, but also demand for its cloud and chips. Anthropic already sells models through Google Cloud’s Vertex AI, and it had already announced in late 2025 that it was expanding its use of Google Cloud TPUs in a deal worth “tens of billions of dollars” th(cnbc.com) Anthropic’s future growth to Google’s TPU roadmap. (anthropic.com) ### Why does the 5 gigawatts matter so much? Because that number is absurdly large. A funding headline can sound like financial engineering. A 5-gigawatt compute commitment sounds like utility-scale industrial planning — because that’s what it is. Anthropic said the new Google and Broadcom agreement covers multiple gigawatts of next-generation(anthropic.com)gement begins next year and could translate into about $200 billion of Anthropic spending with Google over five years. (anthropic.com) ### Where does Ironwood fit in? This is the part where the rumor mill gets fuzzy. Google has an official TPU called Ironwood — its seventh-generation chip, introduced in November 2025. But Google’s Cloud Next event in April 2026 focused on newer eighth-generation TPU 8t and 8i systems. Anthropic’s own announcement did not say “Ironwood” for the multi-gigawatt(anthropic.com)ich TPU generation Anthropic will actually get at that scale. (blog.google) ### Why would Anthropic want this structure? Because frontier AI labs are running into a simple constraint — money alone is not enough if you can’t secure chips, power, and data center capacity years ahead. This kind of package gives Anthropic capital now and a reserved lane on Google’s infrastructure later. Basicall(blog.google)and model training gets more expensive. (anthropic.com) ### Why would Google do it? Google gets a major external customer for its TPUs, a stronger case that its custom silicon can compete with Nvidia-heavy stacks, and a hedge in the model race even if Google’s own first-party models don’t dominate every segment. Anthropic is valuable to Google not just as an equity bet but as a proof point for Google Cloud’s AI in(anthropic.com)ryone else. (cnbc.com) ### What changes for the rest of the market? The bar just moved. The old question was who has the best model. The newer question is who can guarantee enough compute, power, and financing to keep improving that model at frontier scale. That favors hyperscalers and the labs attached to them. Independent model builders now need not just talent and product traction, but balance-sheet-grade infrastructure backing too. (cnbc.com) ### Bottom line? The real news is not that Google might write a giant check. It’s that AI deals are starting to look like power-plant deals — long-dated, capacity-based, and hard for outsiders to match. (anthropic.com)