 Elizabeth Sweeny [LinkedIn](https://www.linkedin.com/in/egsweeny)  François Beaufort [GitHub](https://github.com/beaufortfrancois)  Austin Sullivan [GitHub](https://github.com/a-sully) [LinkedIn](https://www.linkedin.com/in/austin-sullivan)
Published: October 1, 2025
We're excited to announce a significant update to the foundation model in Chrome,
designed to bring powerful, client-side AI capabilities to an even broader
range of users and devices. Listening to your feedback, we're expanding
inference support beyond [GPUs to include CPUs](https://blog.google/technology/ai/difference-cpu-gpu-tpu-trillium/).
## More devices support built-in AI
Our goal is to increase the reach of Gemini Nano. We know that not all users
have devices with GPUs capable of running on-device models efficiently.
By enabling CPU inference on Linux, macOS, and Windows, many more users can
benefit from the performance, privacy, and cost-effectiveness of your website
or web application's built-in AI features.
This expansion is a direct response to the valuable feedback from developers,
especially our early preview program participants, who have requested broader
device compatibility.
### Performance expectations
While CPU support broadens accessibility, inference speeds are typically faster
on devices with capable GPUs. Our focus with this update is to bring built-in AI features to a larger user base.
We continue to work on performance optimization across all hardware, and expect
to make further improvements.
## Your applications work as-is
The best news is that you don't have to change how you interact with built-in AI
APIs in Chrome. Gemini Nano remains consistent across both GPU and CPU
inference. This means:
- No changes to your prompts: Your existing prompt engineering and API calls will continue to work as-is. While response times might differ, the results themselves will remain consistent.
- Consistent API usage: Continue using the [Built-in AI APIs](https://developer.chrome.com/docs/ai/built-in-apis.md.txt) as before.
- Larger user base: Your AI-powered features will function on more devices.
## What's next?
CPU support for Gemini Nano is rolling out in Chrome 140. We encourage you to
test your web applications and extensions to measure their performance across a
variety of devices.
We're thrilled to be making client-side AI available to more users and look
forward to seeing what you build.
Join the early preview program to offer early feedback to new API proposals and
features. [Chrome for Developers](https://developer.chrome.com/blog) blog for more updates.
To share your feedback or file bugs about a specific API's implementation in
Chrome, read the individual API documentation and look for the prompt to file
a bug report.