Google pledges $40B to Anthropic

AI infrastructure is the story here — not just funding. Anthropic did announce a major expansion with Google, but the clean, sourced version is narrower than the viral framing. On April 6, 2026, Anthropic said it signed a new agreement with Google and Broadcom for multiple gigawatts of next-generation TPU capacity, with that capacity expected to come online starting in 2027. Earlier, Anthropic had also said it planned to expand its use of Google Cloud technologies by up to 1 million TPUs in a deal worth “tens of billions of dollars.” (anthropic.com) ### Did Google actually pledge $40 billion? Not in the public material I could verify. Anthropic’s own announcements describe the newer Google-Broadcom arrangement in terms of “multiple gigawatts” of TPU capacity, and the earlier Google Cloud expansion as worth “tens of billions of dollars.” That supports the idea of a huge long-t(anthropic.com) the safest read is: very large, yes — specifically $40 billion, not confirmed in the primary sources I found. (anthropic.com) ### What is Google giving Anthropic? Basically, future compute. The April 6 announcement is about next-generation TPUs, with Broadcom in the mix because Broadcom co-designs Google’s TPU hardware. Anthropic framed the deal as infrastructure for frontier Claude models and for serving fast-growing customer demand. This is less like w(anthropic.com)before everyone else grabs it. (anthropic.com) ### Why do gigawatts matter so much? Because AI scale is turning into an energy-and-datacenter problem. Anthropic said last year that single frontier training runs could require 2 GW data centers in 2027 and 5 GW in 2028. That is an absurd amount of power by normal software standards, but it tells you what these companies think th(anthropic.com) abstract — it is guaranteed, integrated compute at utility scale. (anthropic.com) ### Why is Broadcom in this deal? Because TPUs are not off-the-shelf cloud widgets. Google’s AI stack depends heavily on custom silicon, and Broadcom has long been central to that hardware pipeline. So Anthropic is not just buying cloud instances. It is plugging into Google’s custom chip roadmap — which can matter for performance, cost, and how much capacity can actually be delivered on schedule. (anthropic.com) ### Is Google Anthropic’s only big compute partner? No — and that is part of the bigger shift. Anthropic announced on April 20 a separate expansion with Amazon for up to 5 GW of new compute, including Trainium2 and Trainium3 capacity. It also announced a Microsoft-Nvidia partnership in late 2025 with an initial compute commitment (anthropic.com)of betting everything on one cloud. (anthropic.com) ### So what changed? The important change is that these are no longer ordinary cloud contracts. They look more like power-purchase agreements for intelligence production — long-dated, capacity-focused, and tied to custom hardware roadmaps. That gives model companies more visibility on future costs and deployment, but it also ties the AI race(anthropic.com)s. (anthropic.com) ### Bottom line? The verified story is still big enough without the inflated headline. Anthropic has locked in massive future Google TPU capacity, starting in 2027, and paired it with similarly huge commitments from Amazon and others. The real signal is not one magic dollar figure — it is that frontier AI now runs on reserved gigawatts, not just venture capital. (anthropic.com)