In the global artificial intelligence (AI) competition, Google's tensor processing unit (TPU) is regarded as a key asset. Google Cloud recently launched the seventh-generation TPU "Ironwood", which is specially designed for generative AI inference and supports up to 9,216 liquid-cooled chips. The performance is 10 times higher than the previous generation and the performance per watt is nearly 2 times higher. It will be officially launched in November 2025.
Anthropic announced that it will expand its use of Google Cloud TPUs, expecting to acquire up to 1 million TPUs with a total value of tens of billions of dollars, and provide more than 1GW of computing capacity in 2026. Anthropic's strategy is multi-platform parallelism, using Google TPU, Amazon Trainium and NVIDIA GPU at the same time to ensure flexibility and efficiency in model development.
Google Cloud CEO Thomas Kurian said that the price performance and efficiency of TPU are the main reasons for attracting Anthropic to expand cooperation. Anthropic CFO Krishna Rao also emphasized that customer demand for Claude models continues to grow, and expanding the use of TPU will ensure that the models stay ahead of the curve.
However, the market has growing doubts about whether the AI craze is a bubble. The Bank of England has warned of a possible "sudden correction" in global financial markets, with tech AI companies appearing to be overvalued. Although giants such as Google continue to invest in AI infrastructure, market fragility and energy demand are still issues that cannot be ignored.
The contradiction at the heart of the trillion-dollar AI race Anthropic to Expand Use of Google Cloud TPUs and Services Ironwood: The first Google TPU for the age of inference Expanding our use of Google Cloud TPUs and Services Announcing Ironwood TPUs General Availability and new Axion VMs to power the age of inference Further reading: Google's seventh-generation Ironwood TPU is fully available, expanding its Axion CPU portfolio