Google is now processing 3.2 quadrillion tokens every single month across its entire AI ecosystem, according to CEO Sundar Pichai. He made the announcement during the Google I/O 2026 keynote on May 20.
This staggering number represents a 7x increase from last year.
The volume spans inference across all of Google’s AI-powered surfaces: the Gemini assistant, Search with AI Overviews, YouTube, Workspace, Cloud APIs, and multimodal data processing for images, video, and audio.
For context, Google processed roughly 9.7 trillion tokens per month in April 2024. By May 2025, that jumped to 480 trillion. By October 2025, it approached 1.3 quadrillion. Seven months later, it sits at 3.2 quadrillion.
Pichai noted Gemini now has 900 million monthly active users. AI Overviews serves more than 2.5 billion users globally, and 8.5 million developers are building on Google’s models each month.
To support this scale, Google relies on its custom-built Tensor Processing Units, or TPUs. The company has been designing these chips for AI workloads since 2016.
Google’s stock has more than doubled since last year’s I/O conference, according to Pichai.