Google’s Golden Goose Is Becoming an AI Agent Platform

Gene Munster, Brian BakerMay 22, 2026

Google Search has been under threat ever since AI chatbots changed how people discover information late in 2022. That matters because Search drives about 55% of the company's revenue. At Google's I/O 2026 developer conference this week, it was clear that Google has solved for the chatbot headwind. The company is increasingly turning Search into an AI-native action layer, with agents, commerce rails, and subscriptions. All that means that Search can continue to grow at 10% plus for the next few years (compared to 19% in March, and around 11% on average over the past three years).

Key Takeaways

Search is going agentic, moving from answers and links to actions, tracking, and commerce.

Gemini is becoming the connective tissue across Search, YouTube, Docs, Gmail, Drive, Android, and Cloud.

Google silicon is becoming more real, which lowers the cost to deliver Google Cloud, and helps them gain share.

Agentic Search

The biggest takeaway from I/O is that Google is rebuilding Search around agents.

Liz Reid, head of Search, introduced an intelligent search box that uses multimodal reasoning to reformulate questions and help users get to better answers. The more interesting piece: Google is adding information agents that work in the background 24/7 to track topics as they evolve. Think of Search that never stops.

Classic Search was built around intent capture and link discovery. AI Search is moving toward intent completion. You ask, Search reasons, and agents track your search for changes when you go away.

Google’s biggest risk over the past few years has been people no longer thinking of it as the place where finding info begins. But the last year proved that for most consumers, Search is still habitual, and Google has turned that traffic into chatbot-style answers, which is what users actually want. Information agents are what comes next.

As for monetization, that will look different over time. AI Mode answers may eat into some traditional click volume, but agentic Search opens the door to new ad formats and increases the likelihood of monthly subscription revenue.

The Gemini Operating Layer

AI usage inside Google’s products is growing fast. The amount of AI work Google’s services run every month is up roughly 7x in the past year and about 330x in the past two years. Google now has 13 products with more than a billion users each, plus 8.5M developers building on its models every month.

What we have seen is that Gemini doesn’t have to become the one app that wins every user like ChatGPT. It can become the intelligence layer inside products people already use.

You saw that throughout the I/O keynote. Ask YouTube lets you search inside video, jump to timestamps, and create comparison tables. Docs Live turns a verbal brain dump into a formatted document and pulls in context from Drive and Gmail. Gemini Spark, the headliner, is a personal agent that coordinates across Docs, Gmail, and Drive, running 24/7 on dedicated VMs in Google Cloud so you don’t need your laptop open.

The Custom Silicon Edge

Cloud is the cleaner AI revenue story, with growth up 63% y/y in March, compared to 48% in December. The March growth rate was ahead of Microsoft Azure, up 39%, and AWS, up 28%.

Google is now building its 8th-generation custom silicon, which it unveiled at Cloud Next last month. The new chips lower the cost for Google to deliver Cloud and also lower the cost to users, which has played a role in Cloud’s growing market share.

Today, I estimate less than 5% of Google Cloud’s silicon is custom. As that share grows, the profitability of Google Cloud will inch higher. Long term, I still expect Nvidia to account for the majority (~80%) of Google Cloud’s AI chips. NVIDIA’s performance and software advantage will continue to be appealing for the most advanced workloads.

Disclaimer