Anthropic has expanded its partnership with Google and Broadcom to power its Claude AI models using next-generation TPU chips at massive scale. The new agreement will give Anthropic access to around 3.5 gigawatts of compute capacity, expected to come online starting in 2027. This builds on an earlier deal that already provided large-scale TPU infrastructure through Google Cloud. Broadcom is working along with Google to develop these custom TPU chips, supporting the overall hardware setup needed to run Claude at scale.
The expansion comes at a time when demand for Claude is surging sharply. Anthropic claims its annual revenue run rate has exceeded $30 billion, up from about $9 billion at the end of 2025. Enterprise adoption is accelerating along with this growth. Earlier in the year, the company had reported around 500 business customers spending more than $1 million annually on Claude. That figure has now exceeded 1,000 customers, effectively doubling in less than two months.
To support this growth, the AI giant is significantly increasing its investment in physical infrastructure. The company confirmed that most of the new compute capacity tied to this deal will be located in the United States. This aligns with its previously announced commitment to invest about $50 billion into strengthening domestic AI infrastructure, including data centers, power systems, and high-performance computing clusters.
Anthropic continues to pursue a diversified hardware strategy. Claude models are trained and deployed across a mix of platforms, including chips from Nvidia, Google’s TPUs, and Trainium processors from Amazon. This approach allows the company to assign workloads to the most efficient architecture depending on the task – whether it is large-scale training, fine-tuning, or real-time inference. The result is not just better cost-performance, but also greater system resilience, which is critical for enterprise customers depending on Claude for production use cases.
Even Claude currently stands out as the only frontier AI model available across all three major global cloud platforms. It is accessible through Amazon Web Services via Bedrock, through Google Cloud’s Vertex AI, and through Microsoft’s Azure Foundry. This broad availability gives enterprises flexibility in how they deploy AI.
However, despite the growing role of Google, Amazon continues to remain central to Anthropic’s operations. The company describes Amazon Web Services as its primary cloud provider and training partner, with ongoing collaboration on large-scale initiatives like Project Rainier. Meanwhile, the timing of the latest move becomes critical as recent reports suggest that Anthropic is targeting an October 2026 window for a potential IPO, eyeing a raise of over $60 billion.