The signal: Gemini is the platform now, not the app
Sundar Pichai opened Google I/O 2026 with one slide: 900 million people now use the Gemini app every month, up from 400 million a year ago. The Gemini API processes 3.2 quadrillion tokens per month, 19 billion per minute, 7x more than last May. 375 Google Cloud customers each processed over 1 trillion tokens in the past 12 months. AI Overviews now reach 2.5 billion monthly users in Search; AI Mode, launched a year ago, sits at 1 billion. By the time Pichai handed off the stage, Gemini was not being introduced as a chat app. It was being introduced as the layer Google is building everything around.
The product surface widened in lockstep. Gemini 3.5 Flash arrived 4x faster than competing frontier models at less than half the price, with Gemini 3.5 Pro coming next month. Gemini Omni Flash, a multimodal model that generates video from any input, shipped same-day. Gemini Spark, framed as a “24/7 personal AI agent,” began rollout to Ultra subscribers. Co-Scientist, a multi-agent research partner built on Gemini, launched alongside Gemini for Science, which connects agents to 30+ life-science databases and tools. Google Pics arrived for image creation. Intelligent eyewear was previewed for fall, putting Gemini into ambient hardware that Google has not had since Glass died.
And one layer down, a small announcement that matters more than it looks: OpenAI, Kakao, ElevenLabs, and Nvidia all adopted Google’s SynthID watermark and became C2PA-conformant. Competitors do not adopt each other’s standards casually. Google is now setting AI provenance, not just shipping it.
The take. Three years ago Google was the company that lost the AI race after inventing the transformer. Last year it was the company chasing OpenAI. This week it became the company everyone else is starting to route through. The competitive question is no longer “does Gemini beat GPT or Claude on benchmark X.” That framing is over. The question is “what runs on the Gemini layer,” and the answers Google announced this week are: scientific research, multimodal generation, ambient compute, content provenance. That is platform language, not product language. The closest historical analogue is Apple turning iOS from a phone feature into the platform every consumer app routes through. Pichai’s 900 million users is the same shape as Tim Cook’s “1.4 billion active devices” — a distribution moat that lets you set the rules of the surface, not just compete inside it.
For builders, this changes the bet shape. Building on Gemini is no longer mostly about model quality. It is about access to Google’s distribution surface — Search, Pics, intelligent eyewear, Workspace, life-science partnerships. The model is becoming a commodity inside that surface. The trade-off: Google gets to set the rules of the next AI stack, including provenance standards, agent connectivity defaults, and pricing tiers. For independent builders, this is the moment to plan for consolidation. Pick the surface you want to ride.
Definitions:
- MAU (Monthly Active Users): a standard product metric counting unique users who interacted with the product at least once in a 30-day window.
- Token: the unit of text or media a language model processes. One English word is roughly 1.3 tokens. 3.2 quadrillion tokens/month means ~74 billion tokens per second flowing through Gemini’s APIs.
- Multi-agent: an architecture where several AI agents perceive and act on the same shared environment, coordinating with each other instead of working alone. Co-Scientist is Google’s multi-agent setup for research.
- C2PA (Coalition for Content Provenance and Authenticity): an industry standard for embedding cryptographic metadata into media files to prove who created or edited them and how.
- SynthID: Google DeepMind’s invisible watermarking technology for AI-generated images, audio, and video. Adoption by OpenAI, Kakao, ElevenLabs, and Nvidia makes it an emerging cross-vendor standard.
- Ambient compute: software that runs across always-on devices (glasses, watches, speakers, cars) rather than only on screens you actively open. Google’s intelligent eyewear preview is a bet that Gemini becomes ambient.
Sources: Sundar Pichai’s I/O 2026 keynote, Gemini 3.5 / Gemini 3.5 Flash, Co-Scientist on DeepMind, OpenAI on adopting SynthID and C2PA.
More signals at dropwatch.ai. Want them in your inbox? Subscribe below.