Google Begins Direct Sales of AI Chip TPU to External Customers

Google has shifted to a business model of directly supplying its in-house developed AI semiconductor, TPU, to external customer data centers. Amidst a situation where demand for AI semiconductors exceeds supply, Google has begun offering TPUs in a physical delivery format, a move that signals a change in the AI chip market landscape. TPUs have been available through Google Cloud via a rental model since 2018.

Sundar Pichai, CEO of Alphabet and Google, announced during the Q1 2026 earnings call that the company would begin directly deploying TPUs in specific customers' own data centers due to increasing demand from AI research labs, capital market firms, and high-performance computing (HPC) sectors.

This decision for direct TPU sales is expected to contribute to funding for next-generation silicon research and development and help achieve economies of scale. CFO Anat Ashkenazi projected that some revenue would be generated this year, with full financial contributions starting from next year.

Google unveiled two types of 8th generation TPUs at the 'Google Cloud Next 2026' event on April 22, 2026: the 'TPU 8t' for training and the 'TPU 8i' for inference. These will be offered alongside a variety of computing options, including existing GPUs and its own CPU, 'Axion'.

Key customers that have adopted TPUs so far include Anthropic, Meta, Thinking Machines Lab, Hudson River Trading, and Boston Dynamics. Anthropic officially formalized a contract to adopt one million TPUs in October last year, and Meta has also signed a supply agreement worth billions of dollars.