Anthropic, a leading artificial intelligence developer, has significantly expanded its commitment to Google’s proprietary AI chips and cloud services, securing an immense increase in computing power to train its next generation of models.
The agreement grants Anthropic access to computing potential exceeding one gigawatt. This substantial capacity, equivalent to millions of Google’s Tensor Processing Units (TPUs), is slated for utilization starting in 2026.
This development highlights Anthropic’s deepening strategic partnership with Google. Google is also a major investor in the AI startup.
Anthropic stated that the “price-performance” ratio and overall efficiency of Google’s TPUs were primary factors in the expanded deal. The company noted positive past experiences training its Claude AI models on the chips.
Google originally developed TPUs for its own internal artificial intelligence workloads. These specialized chips are now available through Google Cloud, serving as an alternative to highly sought-after semiconductors from NVIDIA.
The global demand for high-performance AI chips has surged. This has fueled competition among major tech companies to offer powerful and cost-effective computing solutions.
As part of the arrangement, Google will also provide general cloud services to Anthropic.