Google looks behind in AI. Its biggest advantage is already in half the world's pocket

Google has lost the narrative, and that matters

For years, it was almost impossible to discuss artificial intelligence without ending up at Google. The Transformer was born there, DeepMind turned scientific problems into headlines, and the company had money, data, talent and its own chips. Today the picture feels different: OpenAI sets the agenda, Anthropic has won over a large share of developers, and Chinese labs release cheaper models at an uncomfortable pace.

The latest bruise is Gemini 3.5 Pro's delay. Google promised it for June at I/O, but a Bloomberg report carried by Reuters says it remains months behind its internal schedule, particularly because of coding performance. That does not make Gemini a failure. It does reinforce the image of a giant company chasing rivals that can decide and ship with less friction.

Nine hundred million users change the picture

The AI race is often reduced to whichever model tops a benchmark for a week. Through that lens, Google can look behind. Distribution tells a very different story: the company says the Gemini app has passed 900 million monthly users, AI Overviews reaches more than 2.5 billion and AI Mode in Search has already crossed one billion.

Those are Google's own figures, not an independent audit, but the scale matters. ChatGPT may release a better model and Claude may write better code; neither controls Android, Gmail, Maps, YouTube, Chrome, Drive and the world's most widely used search engine at once. Google has thirteen products with more than a billion users. It does not need everyone to install a new app when it can place Gemini inside tools people already open every morning.

Its best product may be an AI that stops looking like an app

The genuinely hard-to-copy move is not another chatbot. It is a Gemini that understands the email that just arrived, finds the right document, remembers the video you watched, drafts a reply and warns you when to leave based on traffic—not as ten separate demos, but as one continuous experience.

Google is already moving that way with agents in Search and Gemini Spark, an assistant designed to work for hours in the background. It is the same race being run by GPT-5.6 and the new generation of work agents, except Google begins with permissions, accounts and services its rivals must borrow. The challenge is not collecting the pieces. It is joining them without creating a confusing, invasive product that acts where nobody asked it to.

Search is a huge advantage—and a trap

Distribution does not guarantee victory. Microsoft owned the desktop and lost mobile. Google owned search and still failed to stop ChatGPT changing how millions of people ask questions. The company also faces a conflict its rivals feel less sharply: answers inside Search can reduce clicks, anger publishers and disrupt the advertising business paying for much of the investment.

Google expects to spend roughly $190 billion on infrastructure this year. That is not the budget of a company preparing to surrender. Spending more, however, does not guarantee a product people want. The question is no longer whether Gemini wins one table, but whether Google can turn its enormous advantage into a coherent and useful experience.

For now, it is losing the conversation and some of the speed. Writing it off would still mean watching a 100-metre sprint while Google may be trying to build the operating system for AI. It arrived late to the defining chatbot. It cannot afford to arrive late to the layer connecting everything else.