Google Cloud TPU Business Doubles in Size Amid AI Boom
Google Cloud's TPU (Tensor Processing Unit) business is thriving, according to CEO Thomas Kurian. In an interview at the Goldman Sachs Communacopia Technology Conference, Kurian revealed that Google's TPU accelerator chip business has doubled in size compared to its hyperscaler competitors like AWS and Microsoft.
The return on investment for Google TPUs is faster than GPUs (Graphics Processing Units), with a payback period of less than two years for AI servers. In comparison, the payback period for Google's own silicon is half that of GPUs.
Kurian attributed this success to Google's diverse range of TPU accelerators, which cater to different types of models and workloads. He emphasized that one type of accelerator cannot solve everything, citing sparse and dense models as examples.
The CEO also highlighted the popularity of Google Cloud's flagship agentic AI platform, Gemini Enterprise, which is used by over 90% of Fortune 100 companies and thousands of small businesses. Gemini Enterprise allows users to build custom AI agents for various use cases, making it an attractive option for enterprises.