Detailed Analysis
Anthropic has secured what the company describes as its most significant compute commitment to date, forging expanded partnerships with Google and Broadcom to access approximately 3.5 gigawatts of next-generation TPU capacity, with the bulk of that infrastructure expected to come online in 2027. The deal extends an earlier October 2025 agreement between Anthropic and Google Cloud — which had already locked in over one gigawatt of TPU capacity — and now incorporates Broadcom as a critical hardware intermediary, supplying custom TPUs for Google's future chip generations as well as networking components for AI racks through 2031. The arrangement routes a portion of that capacity to Anthropic contingent on its commercial performance, a structuring choice that ties infrastructure scale directly to revenue milestones. CFO Krishna Rao characterized the commitment as necessary to keep pace with what the company describes as exponential customer growth and to continue advancing frontier AI development.
The scale of these commitments reflects Anthropic's rapidly accelerating commercial trajectory. The company's annualized run-rate revenue has reached $30 billion, and its base of enterprise customers more than doubled — from 500 in February 2026 to over 1,000 — within a matter of months. Claude is now available across all three major cloud hyperscalers: Amazon Web Services via Bedrock, Google Cloud via Vertex AI, and Microsoft Azure via Foundry, a multicloud distribution strategy that broadens Anthropic's addressable market while creating redundancy in deployment channels. Amazon remains Anthropic's primary cloud and training partner through Project Rainier, and the company continues to train Claude across a diversified hardware stack — including Google TPUs, Nvidia GPUs, and AWS Trainium chips — both for performance optimization and supply-chain resilience.
The infrastructure expansion sits within a broader strategic and geopolitical context. The majority of the new capacity will be located in the United States, consistent with the $50 billion commitment to American AI infrastructure that Anthropic announced in November 2025. That framing positions the company as a participant in the domestic industrial policy push around AI compute sovereignty — a calculation that carries both regulatory and reputational significance as governments increasingly scrutinize where frontier AI systems are trained and hosted. The multi-gigawatt scale also signals that the resource requirements for training and running frontier models have grown dramatically, making large, long-dated infrastructure deals with hyperscalers and chip suppliers a near-prerequisite for competing at the top tier of AI development.
Analysts view the Google-Broadcom-Anthropic arrangement as emblematic of a structural shift in how AI competition is playing out. The differentiating variables are no longer limited to model architecture or benchmark performance; they increasingly include systems integration, power procurement, and chip interconnectivity. This dynamic advantages firms with deep hyperscaler relationships, as those partnerships provide preferential access to scarce compute at a time when global demand for AI training infrastructure far outpaces supply. For Broadcom, the financial stakes are substantial: Mizuho estimates the firm will generate $21 billion in AI-related revenue in 2026 and $42 billion in 2027, with the Anthropic relationship representing a significant contributor. The deal thus illustrates how the AI supply chain has become deeply interlocked — with model developers, cloud providers, and semiconductor companies bound together by long-horizon contractual commitments that reflect the capital intensity and strategic urgency of the current moment in AI development.
Read original article →