Google is accelerating the transformation of its self-developed tensor processing unit (TPU) from an internal tool into a commercial AI chip sold externally, directly challenging Nvidia's dominance in the AI hardware market. The technology giant has made TPU a core component of its AI supercomputers, and through its partnership with Broadcom, Google's TPU business has expanded to provide complete AI infrastructure solutions to external customers such as Anthropic.
In the current AI chip market landscape, Nvidia still dominates with about 86% of data center chip revenue. However, Google’s TPU is leveraging this landscape with its cost and system advantages. According to reports, Google’s self-developed TPU can process AI workloads at a cost that is 30% lower than competing processors. This advantage is particularly significant in large-scale deployment.
At a strategic level, Google has completed the transformation from internal use only to full commercialization in recent years. Previously, Google reached an agreement with Anthropic, which will deploy up to 1 million Google seventh-generation TPUs for training its Claude model. The deal is said to be Google's first time competing with Nvidia as a direct hardware supplier, marking a fundamental shift in its TPU strategy.
At the same time, Google has released an eighth-generation TPU specifically optimized for training tasks and inference tasks, and plans to launch it later this year. About 75% to 80% of the company's current TPU production capacity is still used for internal business, but analysts predict that its TPU production capacity will further expand, with annual TPU output expected to reach 5 million pieces by 2027.
Analysts believe that Google TPU’s external sales strategy has been regarded as the most structural threat to Nvidia’s GPU dominance in the AI chip market. Although Google still faces challenges in the software ecosystem, its customer success stories and growing external demand are highlighting its feasibility as an alternative to Nvidia.
