This momentum is a direct result of our decade-long full-stack approach to AI, our unique differentiator. The foundational layer of the stack is our unrivaled AI infrastructure.
This includes 10 million kilometers of terrestrial and subsea fiber connecting over 30 data centers and over 40 Cloud regions around the world. Our recent acquisition of Intersect will help us come to market even faster.
We pride ourselves on building efficiently and sustainably, and supporting local communities through job creation and workforce development programs.
We offer a wide variety of compute options – from our own TPUs and Axion CPUs to the latest NVIDIA GPUs, which are a core part of our AI accelerator portfolio.
It’s been 10 years since we introduced our first commercial TPU. Today, they power Gemini model training and serving at scale across our APIs and biggest products like Search.
Our eighth-generation TPUs, 8i and 8t, offer a new dual-chip design specifically for the massive training and inference needs of the agentic era. We continue to drive tremendous efficiency gains; for example, in 2025, we reduced Gemini serving costs by 78%.
Security is the critical second layer of our full stack, built directly into our foundational code. With the rapid adoption of AI, protecting data and privacy grows more complex every day. Our secure-by-default architecture automatically blocks 10 million spam emails every minute, and protects billions of users and customers across our broad portfolio. We are proud to keep more people safe online than anyone else in the world.
For our enterprise customers, we launched Google AI Threat Defense. This is a new security platform that combines the strengths of Gemini and other frontier models, along with Wiz, CodeMender and Mandiant, to stop AI-powered threats before they impact our customers.
The next layer is our world-class research teams, led by Google DeepMind. These teams are the engine of our entire AI stack, from inventing the Transformer architecture to developing our highly capable Gemini models.
Research teams are also pioneering advances in science, medicine and climate. That includes WeatherNext and AlphaFold, which was awarded the Nobel Prize in chemistry.
That brings us to our fourth layer: models and tooling. We offer the most extensive model portfolio in the world, and we continue to stay on the cutting edge in reasoning, multimodality and cost.
We recently introduced Gemini 3.5 with improvements focused on agentic coding, long-horizon tasks and real-world capabilities. Gemini 3.5 Flash is now available, and we expect Gemini 3.5 Pro will be coming in June.
Beyond the Gemini series, our generative media models are incredibly popular. This includes Nano Banana, which people have used to generate more than 50 billion images to date. And we were excited to unveil our latest Omni model at I/O.
One way to measure growth is tokens, the fundamental units of data our models process. Two years ago, we were processing 9.7 trillion tokens a month. Today, that number has jumped to 3.2 quadrillion, a more than 300 times increase across our surfaces.
The speed of adoption from enterprises and developers is incredible:
- More than 8.5 million developers are now building new apps and experiences with our models each month.
- Our model APIs are now processing approximately 19 billion tokens per minute.
- And over the past 12 months, over 375 Cloud customers have each processed more than one trillion tokens.
Tooling is another important part of this layer. Antigravity, our agentic development platform, launched six months ago and we have millions of developers building with it. We recently expanded Antigravity to a unified platform for builders to work with agents. This includes a standalone desktop app that delivers the first major step towards our vision of an independent, agent-focused surface. And for the first time, we are bringing the Antigravity ecosystem to Google Cloud customers.
The Antigravity coding harness has accelerated how we build internally. Every few weeks, we’re doubling the number of tokens we’re processing across our developer tooling. Recently, we reached more than three trillion tokens a day. This scale creates a powerful feedback loop to improve our current and future models.
Finally, the fifth layer of our stack is our unmatched portfolio of products and platforms used by billions of people every day.
