Anthropic is going all in on compute with Google and Broadcom

Anthropic just announced a new agreement with Google and Broadcom that sounds like the kind of thing you’d hear from a company that’s growing faster than it can build.

They’re committing to multiple gigawatts of next-generation TPU capacity, with the first tranche coming online in 2027. That’s a lot of compute, and it’s all going toward powering Claude and handling the insane demand they’re seeing from customers.

Let’s talk numbers, because they’re the real story here.

Anthropic’s run-rate revenue has hit $30 billion. That’s up from about $9 billion at the end of 2025. In less than two years, they’ve more than tripled. And the customer base isn’t just growing—it’s getting deeper. Back in February during their Series G announcement, they said over 500 business customers were spending more than $1 million annually. Today that number is over 1,000. Doubled in under two months. That’s not growth, that’s a stampede.

Krishna Rao, Anthropic’s CFO, called this their “most significant compute commitment to date.” He’s not wrong. The scale here is hard to wrap your head around—gigawatts of TPU capacity isn’t something you see every day.

Most of this new compute will be built in the United States, which is an expansion of the $50 billion commitment they made back in November 2025 to strengthen American computing infrastructure. I like that they’re putting their money where their mouth is on that front.

What I find interesting is the hardware diversity angle. Anthropic trains and runs Claude on AWS Trainium, Google TPUs, and NVIDIA GPUs. That’s three very different architectures. Most AI companies pick a lane and stick with it. Anthropic is deliberately spreading the load, matching workloads to whatever chip handles them best. That’s smart engineering, and it also means they’re not locked into any single vendor. If one supply chain hiccups, they can shift capacity. Customers who depend on Claude for critical work benefit from that resilience.

Amazon remains their primary cloud provider and training partner, and they’re still working on Project Rainier together. But Claude is also available on Google Cloud (Vertex AI) and Microsoft Azure (Foundry). They’re the only frontier AI model you can run on all three major clouds. That’s a flex, and it’s a practical one for enterprises that have multi-cloud strategies.

This partnership with Broadcom isn’t new—they’ve been working together for a while. But deepening it at this scale signals that Anthropic sees TPUs as a core part of their future, not just a side experiment.

The 2027 timeline is worth noting. That’s not next quarter. That’s planning two years out for infrastructure that doesn’t exist yet. It tells you how seriously they’re taking the compute arms race, and how confident they are that demand will keep climbing.

I don’t usually get excited about infrastructure announcements, but this one matters. It’s a concrete signal that Anthropic isn’t just riding the AI wave—they’re building the foundation to survive whatever comes next.