Anthropic just announced a major expansion of its compute partnership with Google and Broadcom, committing to multiple gigawatts of next-generation TPU capacity that’s expected to come online starting in 2027. This isn’t just a small bump in capacity — we’re talking about enough compute to power frontier Claude models and keep up with what they describe as “extraordinary demand” from customers worldwide.
Krishna Rao, Anthropic’s CFO, framed this as a continuation of their “disciplined approach to scaling infrastructure.” I’ll buy that, but let’s be real: when your revenue run-rate jumps from $9 billion at the end of 2025 to over $30 billion in early 2026, you’re not being disciplined — you’re sprinting to keep the lights on. The numbers are staggering. Back in February during their Series G announcement, they had 500 business customers each spending over $1 million annually. That number has now doubled to over 1,000 in less than two months. That’s not growth, that’s a land rush.
What’s interesting here is the hardware strategy. Anthropic is keeping its options wide open. They train and run Claude on AWS Trainium, Google TPUs, and NVIDIA GPUs, matching workloads to the best chip for the job. This multi-platform approach gives them both performance flexibility and resilience — if one supplier hits a snag, they can shift load. Amazon remains their primary cloud provider and training partner, and they’re still working with AWS on Project Rainier. But this new deal with Google and Broadcom signals that they’re not putting all their eggs in one basket.
Most of the new compute will be sited in the US, building on their November 2025 commitment to invest $50 billion in American computing infrastructure. That’s a smart political move too, given the current climate around AI and national security.
One thing that stands out: Claude remains the only frontier AI model available on all three major cloud platforms — AWS Bedrock, Google Cloud Vertex AI, and Microsoft Azure Foundry. That’s a genuine differentiator. OpenAI is locked into Azure, and Google’s Gemini is obviously Google-only. Anthropic is playing the field, and it’s paying off.
I’m curious to see how this plays out with Broadcom. They’re not usually the first name that comes to mind in AI chips, but they’ve been quietly building custom ASICs and networking gear for hyperscalers. This deal suggests they’re becoming a serious player in the AI infrastructure supply chain.
The real question is whether this compute will arrive fast enough. 2027 feels like a long way off when demand is doubling every couple of months. But if anyone can manage that kind of scaling curve, it’s Anthropic — they’ve been doing it since day one.
Comments (0)
Login Log in to comment.
Be the first to comment!