At Google I/O 2026, CEO Sundar Pichai opened the keynote by declaring the start of the agentic Gemini era. The headline numbers set the tone: Google surfaces now process over 3.2 quadrillion tokens a month — a 7x jump year over year — with 8.5 million developers building on Google models monthly.
Full-Stack Momentum
Pichai walked through adoption across the stack:
- 13 products with more than 1 billion users, five of those past 3 billion
- AI Overviews now at 2.5 billion monthly active users
- AI Mode crossed 1 billion monthly active users in its first year
- The Gemini app grew from 400 million to over 900 million MAU
- Over 50 billion images generated with Nano Banana models
Gemini 3.5 Flash
The centerpiece was Gemini 3.5 Flash, available today across products and APIs. Google says it beats 3.1 Pro across almost all benchmarks and is roughly 4x faster at output tokens than other frontier models, positioning it as the cheapest route to frontier-adjacent intelligence — shifting just 80% of workloads to Flash could save top companies over $1 billion a year. A further-optimized variant is 12x faster inside Antigravity.
Agents Everywhere
The agentic push is the core of this year’s event:
- Gemini Spark, a 24/7 personal agent running on dedicated Google Cloud VMs, starts rolling out to trusted testers this week
- Daily Brief, an out-of-the-box digest agent, is coming to the Gemini app
- Information agents in Search arrive this summer for Google AI Pro and Ultra subscribers
- Android Halo brings agent task progress to the status bar later this year
Infrastructure: TPU 8t and 8i
Google unveiled its 8th-generation TPU split into two specialized chips: TPU 8t for large-scale pretraining, nearly 3x the raw compute of the prior generation and scalable past 1 million TPUs across multiple data-center sites; and TPU 8i for inference. Both deliver up to 2x better performance-per-watt. Total capital expenditure is expected to reach $180–190 billion this year, about six times 2022 spending.
Transparency and SynthID
Pichai highlighted SynthID’s growth — over 100 billion watermarked images and videos plus 60,000 years of audio — and announced that OpenAI, Kakao, and Eleven Labs are adopting SynthID, alongside expanded Content Credentials verification in Search and Chrome.
What This Means
I/O 2026’s through-line is that agents are no longer demos — Gemini Spark, Antigravity 2.0, and information agents are shipping products. Google is turning its scale advantage (13 billion-user products, 1-million-TPU clusters) into a distribution moat for agentic AI that rivals have to answer, not match.