Anthropic just announced a new agreement with Google and Broadcom that’s frankly staggering in scale: multiple gigawatts of next-generation TPU capacity, expected to come online starting in 2027. If you’ve been following the AI infrastructure arms race, this is the kind of number that makes you sit up.
For context, a gigawatt is roughly the output of a large nuclear reactor. Multiple gigawatts of dedicated compute for one company’s models is not just “big” — it’s the kind of commitment that reshapes supply chains. This is Anthropic’s most significant compute deal to date, and it shows they’re betting hard that demand for Claude isn’t a bubble.
And the revenue numbers back that up. Anthropic’s run-rate revenue has hit $30 billion, up from around $9 billion at the end of 2025. That’s a 3x growth in roughly a year, which is bonkers by any standard. Even more telling: back in February during their Series G announcement, they said over 500 business customers were spending over $1 million annually on Claude. That number has now doubled to over 1,000 in less than two months. That’s not just growth — that’s a land grab.
Krishna Rao, Anthropic’s CFO, framed it as a “disciplined approach to scaling infrastructure,” which is a polite way of saying they’re trying not to run out of compute while demand explodes. Fair enough.
What I find interesting is the hardware strategy. Anthropic is running Claude on AWS Trainium, Google TPUs, and NVIDIA GPUs — a multi-platform approach that gives them flexibility to match workloads to the best chip. That’s smart, especially given the GPU shortage we’ve seen over the last couple of years. It also means they’re not locked into any single vendor, which is probably a good hedge.
Amazon remains their primary cloud provider and training partner, and they’re still working on Project Rainier. But this new deal with Google and Broadcom is clearly a major expansion of their November 2025 commitment to invest $50 billion in US computing infrastructure. Most of this new compute will be sited in the US, which plays well with the current political climate around AI sovereignty.
One thing that stands out: Claude is still the only frontier AI model available on all three of the world’s largest cloud platforms — AWS Bedrock, Google Cloud Vertex AI, and Microsoft Azure Foundry. That’s a genuinely unique position, and it gives enterprise customers a lot of flexibility. If your company is already deep in one cloud ecosystem, you don’t have to migrate to use Claude.
Of course, the elephant in the room is whether this kind of spending is sustainable. $30 billion run-rate is impressive, but the compute costs at this scale are enormous. Anthropic is essentially making a bet that the demand curve keeps going up, and that they can capture enough of the enterprise AI market to justify the infrastructure spend. Given the pace of adoption, I wouldn’t bet against them.
But it’s worth noting that this is a race with no finish line. OpenAI, Google DeepMind, and others are all making similar moves. The question isn’t whether Anthropic can build the compute — it’s whether they can keep Claude ahead of the pack while doing it.
Comments (0)
Login Log in to comment.
Be the first to comment!