Anthropic and compute scale
Beyond models, Anthropic’s commercial footprint is growing rapidly — the firm is reported to have a roughly $30bn annualised run‑rate and is scaling large enterprise customers spending $1M+ annually. The same coverage links Anthropic’s growth to infrastructure deals that secure gigawatts of compute capacity, highlighting that access to vast compute is now a gatekeeper for frontier competitiveness. That dynamic matters because compute partnerships and capital intensity are becoming as decisive as model quality in who wins the enterprise AI market. (fortuneindia.com)
Anthropic says its revenue run rate has passed $30 billion, up from about $9 billion at the end of 2025, and more than 1,000 business customers now spend over $1 million a year on Claude. That is not a consumer app story anymore; it is a big-company software budget story. (anthropic.com) (finance.yahoo.com) At the same time, Anthropic signed a new deal with Google and Broadcom for multiple gigawatts of next-generation Tensor Processing Unit capacity starting in 2027. A gigawatt is power-plant scale, so the company is now talking about compute the way utilities talk about electricity. (anthropic.com) (googlecloudpresscorner.com) That pairing is the whole story: customers are buying Claude fast enough that Anthropic is locking in years of hardware before the machines are even online. In plain English, it is pre-booking the factory floor before the next rush arrives. (anthropic.com) (bloomberg.com) Anthropic is not building alone. Amazon Web Services put another $4 billion into the company in November 2024, bringing Amazon’s total investment to $8 billion, and Anthropic said Amazon Web Services remains its primary cloud provider and training partner. (anthropic.com) (aboutamazon.com) (anthropic.com) So Anthropic now sits in an unusual position: Amazon is its primary training partner, Google is supplying large new Tensor Processing Unit capacity, and Claude is sold on Amazon Bedrock, Google Vertex AI, and Microsoft Azure Foundry. The model company is starting to look like a tenant with storefronts in every biggest mall. (anthropic.com) (aws.amazon.com) The reason this is happening is simple: frontier artificial intelligence is no longer just a software contest. Training and serving the best models now depends on chips, data centers, networking gear, electricity, and long contracts with the few firms that control them. (anthropic.com) (crn.com) Broadcom’s role shows how deep the stack goes. Anthropic is not just buying cloud time from Google; Broadcom is tied into the custom chip and networking layer that helps turn those future data centers into usable model capacity. (anthropic.com) (crn.com) The enterprise numbers explain why the infrastructure deals can be this large. If 1,000 customers are already spending more than $1 million a year, and that count has more than doubled since February, Anthropic can justify buying compute years ahead because demand is no longer hypothetical. (finance.yahoo.com) (anthropic.com) That changes what “winning” looks like in artificial intelligence. A better model still matters, but so do balance sheets, chip access, cloud alliances, and the ability to secure enough electricity-backed compute to keep the service running when big customers show up. (anthropic.com) (forbes.com) Anthropic’s update reads like a revenue announcement on the surface, but the bigger clue is the shopping list underneath it: Amazon money, Google Tensor Processing Units, Broadcom hardware, and contracts stretching into 2027. In this market, the frontier is not just the model in the demo; it is the power, chips, and cloud capacity behind the curtain. (anthropic.com) (aboutamazon.com)
