source: google ai: i/o 2026: welcome to the agentic gemini era
level: technical
google ceo sundar pichai outlined a year of rapid ai growth at io 2026. token processing across google surfaces jumped to over 3.2 quadrillion per month, a sevenfold increase from last year. over 8.5 million developers now build with google models monthly, and more than 375 cloud customers each processed over a trillion tokens in the past year. ai overviews in search reached 2.5 billion monthly users, and ai mode surpassed 1 billion. the gemini app doubled its users to over 900 million, with daily requests up seven times.
new product features bring conversational ai to more services. ask youtube lets users jump to relevant video segments based on questions. docs live enables voice-driven document creation by verbally describing content. gemini spark is a personal ai agent that runs 24/7 on cloud virtual machines, performing tasks in the background. it will integrate with google tools and third-party services via mcp, and later operate in chrome. android halo will show live agent progress. spark begins rolling out to testers this week.
infrastructure and model updates support this scale. google expects $180-190 billion in annual capex, up from $31 billion in 2022. custom tpu 8 chips use dual architectures for training and inference, with tpu 8t enabling distributed training across over a million tpus. gemini 3.5 flash offers frontier performance at lower cost, with output four times faster than comparable models. gemini omni flash generates video from any input. synthid watermarking expands to search and chrome, with openai, kakao, and eleven labs adopting it.
why it matters: faster, cheaper models and agentic tools can reduce compute costs and enable new ai applications for developers and enterprises.
source: google ai: i/o 2026: welcome to the agentic gemini era