• 2 min read
Chrome may need 20 GB for local AI
Chrome may reserve about 20 GB for local AI models, updates, and supporting components. Users can disable on-device AI.

Image: ITzine
Google’s updated Chrome documentation says the browser may automatically download local AI models when a computer meets several conditions, including compatible hardware, an unrestricted internet connection, and roughly 20 GB of free storage.
That requirement is considerably larger than the model itself. According to iXBT, Gemini Nano needs approximately 4 GB, but Google’s estimate also covers model updates and the components required to run related features. Chrome uses these on-device models for some generative capabilities, although the documentation does not identify every feature covered by the download system.
How Chrome decides whether to download models
Chrome first checks the device’s hardware capabilities. If the computer fails that check, the browser will not download the local models. If it passes, Chrome can use the available internet connection to fetch the files automatically, provided the connection has no traffic limits.
The documentation does not say when this behavior will reach every Chrome installation, which hardware configurations qualify, or whether the 20 GB figure is a hard minimum or a reserved allowance. Those omissions make it difficult to predict the impact on a particular PC, especially where storage is limited.
Disabling on-device AI
Users can turn off the On-device AI setting in Chrome. Doing so removes models already downloaded by Chrome, disables associated features, and prevents the browser from automatically fetching new files.

Recommended reading
Kimi K3 reportedly bypassed AI safety sandbox via GitHub
Google warns against deleting the model files manually. Manual removal does not disable Chrome’s download mechanism, so the browser may simply download them again. The setting applies separately to each user and device; on managed computers, Chrome Enterprise policies can control it for administrators.
The trade-off is straightforward: local AI keeps processing on the computer but can reserve about 20 GB for a model ecosystem whose core Gemini Nano model reportedly occupies only 4 GB. Until Google publishes clearer hardware, rollout, and storage details, the opt-out is the more practical choice for users who cannot spare that space.
AI Editor
Ava covers the rapidly evolving world of artificial intelligence, from foundational models and research labs to the real-world economics of intelligence. With a background in computational linguistics, she cuts through the hype to find out what actually works. She firmly believes that benchmarks are just marketing until reproduced in the wild.
via ITzine


