Anthropic Signs Largest Compute Deal with Google and Broadcom for TPU Capacity Starting 2027

Anthropic, an artificial intelligence company, announced a major compute deal with Google and Broadcom. The agreement provides multiple gigawatts of Tensor Processing Unit (TPU) capacity, marking Anthropic's largest such arrangement to date. Delivery of the capacity is scheduled to start in 2027.

The partnership builds on existing relationships between the companies. Google has previously invested in Anthropic, and Broadcom supplies semiconductor technology used in TPUs. This deal expands access to specialized hardware for training and running AI models.

The agreement highlights increasing demand for computational resources in the AI sector.

Data centers require substantial electricity and grid connections to operate at scale. According to CoinDesk, this demand contributes to broader competition for affordable power and infrastructure, including from Bitcoin mining operations that also rely on energy-intensive facilities. Bitcoin miners often locate in areas with low-cost electricity to maximize profitability.

AI compute needs similarly prioritize regions with reliable grid access. As both industries grow, potential overlaps in resource allocation could affect availability and costs.

expansion aligns with the rapid advancement of AI technologies.

The company focuses on developing safe and reliable AI systems, with TPUs enabling efficient processing for large-scale models. Starting in 2027, the new capacity will support Anthropic's ongoing research and deployment efforts. Stakeholders in energy and technology sectors may monitor how such deals influence power markets.

Regulatory bodies and utilities could face pressures to expand grid capacity. The agreement underscores the intersection of AI development and energy infrastructure planning. No specific details on the exact volume of gigawatts or financial terms were disclosed in the announcement.

Further updates may provide clarity on implementation timelines and technical specifications.