Sundar Pichai walked onto the stage at Shoreline Amphitheatre on Tuesday and made the case that Google’s moment in artificial intelligence has arrived. The company’s platforms now process 3.2 quadrillion tokens per month, the chief executive said at Google’s annual I/O developer conference, a sevenfold increase from a year earlier. The Gemini app, he said, now has more than 900 million monthly active users across 230 countries and 70 languages, up from 400 million when the previous I/O was held in May of last year.
The numbers were the backdrop for a product lineup built around a single idea: agents. Google announced Gemini 3.5 Flash, the next generation of its fast, widely deployed model, and introduced Antigravity, a platform for building agentic applications. It also showed off its first personal agents, consumer-facing systems designed to carry out multi-step tasks on a user’s behalf, from planning trips to managing inboxes.
“Gemini 3.5 and Antigravity are opening a new world of agents and agentic capabilities,” Pichai said. “We have brought agents to developers and enterprises. Now we are focused on bringing the power of agents to consumers, safely and reliably, so that it works for everyone.”
The company’s ambitions were stated in even bolder terms by Demis Hassabis, the chief executive of Google DeepMind. “I think, when we look back on this time, we will find that we were standing one step before the Singularity,” Hassabis told the audience, referring to the hypothetical point at which machines surpass human intelligence.
The conference represents Google’s answer to the challenge that OpenAI posed three years ago, when ChatGPT turned conversational AI into a mainstream product. Google responded with Gemini, and the app’s growth to 900 million users has made it the largest consumer AI product in the world, even as OpenAI’s ChatGPT has continued to grow. The two companies are now racing to turn chatbots into agents that do things rather than merely answer questions.
The stakes extend beyond the apps themselves. Google controls the distribution channels, through Android, Search, Chrome, and Workspace, and its new agent products are designed to live inside those surfaces. A personal agent that can act inside Gmail or Calendar, or that sits in the browser and handles tasks across sites, gives Google a structural advantage that standalone apps cannot match. The company’s challenge is proving that the agents work reliably enough for consumers to trust them with real tasks.
The token figures underline the scale of the infrastructure bet. Processing 3.2 quadrillion tokens per month requires a data-center fleet that few organizations on earth could assemble, and Google has been spending tens of billions of dollars a year on AI capacity. Pichai’s numbers, repeated without qualification, appeared designed to tell investors that the spending is producing usage, and that the usage is growing faster than the costs.
Analysts said the consumer push is the right response to the market’s skepticism about AI monetization. “The question was always whether AI usage would stay concentrated among developers and enterprise buyers,” one analyst said. “Google’s numbers suggest the consumer base is now the bigger story, and agents are the product that could keep them engaged.” The analyst cautioned that agent products carry higher expectations, and that failures, a flight booked to the wrong city or an email sent to the wrong person, would be punished harshly by users.
For developers, the conference’s message was that building on Google’s stack now means building with agents as the default unit of software. Antigravity is positioned as the bridge between the models and the real world, handling the tool use, memory, and multi-step reasoning that agents require. Early partners described the platform as closer to a runtime environment than a model API, a sign that Google wants to own the layer where the next generation of applications is built.
The company’s rivals are moving in the same direction. OpenAI has shipped its own agent tools, and Microsoft has been weaving agent technology through its enterprise software. What distinguishes Google’s approach is distribution: no other AI company can put an agent in front of a billion users through its existing products in a single update cycle.
The I/O announcements will be tested in the coming months, as consumers try the new agents and developers decide whether Antigravity justifies the switch. Pichai’s 900 million figure gives Google a base that no competitor matches. Hassabis’s Singularity remark, whatever its scientific merits, gave the conference a line that will be quoted until next year’s event. Between those two poles, Google is betting that the era of agents belongs to the company with the biggest audience and the deepest infrastructure, and this week it argued, with numbers, that it has both.
