Google says 75% of cloud customers use AI

Google Cloud said nearly 75% of its customers now use its artificial intelligence products, turning AI from a pilot project into a mainstream cloud feature. (cloud.google.com) The company disclosed the figure at Google Cloud Next 2026 in Las Vegas on April 22, alongside new eighth-generation Tensor Processing Units and the Gemini Enterprise Agent Platform for building and managing AI agents. (cloud.google.com) (blog.google) Google said 330 cloud customers processed more than 1 trillion tokens over the past 12 months, and 35 processed more than 10 trillion tokens with its models. It also said its first-party models now handle more than 16 billion tokens per minute through direct customer application programming interface use, up from 10 billion last quarter. (cloud.google.com) (blog.google) A token is a chunk of text a model reads or writes, so token counts are a rough measure of how much AI work customers are actually running. Google used those figures to argue that enterprise demand has moved beyond experiments and into steady production workloads. (blog.google) (computerweekly.com) The hardware push is part of that message. Google introduced separate TPU 8t and TPU 8i systems, with the company positioning one for training large models and the other for inference, the step where a trained model answers prompts in real time. (forbes.com) (blog.google) Google also rebranded its enterprise agent tooling around Gemini, adding controls for building, governing and optimizing agents across systems. Workspace Intelligence, another launch at the event, extends those AI tools into Google’s office software. (blog.google) (crn.com) Alphabet is backing the cloud push with unusually large spending. In February, the company said it expects 2026 capital expenditures of $175 billion to $185 billion, versus $91.4 billion in 2025, and Sundar Pichai reiterated that plan this week. (cnbc.com) (economictimes.indiatimes.com) Pichai said just over half of Google’s machine learning compute investment in 2026 is expected to go to inference, not training. That is another sign the company expects customers to spend more on serving AI systems at scale than on building prototypes. (blog.google) The immediate test is whether Google can turn that usage into durable cloud revenue while competing with Microsoft and Amazon for the same corporate AI budgets. For now, Google is arguing that the market has shifted from proving AI can work to proving it can run reliably and cheaply enough for everyday business use. (cnbc.com) (computerweekly.com)