Chrome secretly takes up 4GB of hard drive space: the hidden cost of Gemini Nano’s local AI model

Core conclusion

After turning on the AI function, Google Chrome will automatically download a Gemini Nano local AI model file (weights.bin) of about 4GB, which is used for in-browser AI functions such as fraud detection, writing assistance, and auto-filling. Because the download process happens silently in the background, many users only discover that several gigabytes are occupied when they check their hard drive space.

Impact on ordinary users: If your Chrome has the Gemini AI function turned on, it will most likely occupy about 4GB of disk space. The good news is that you can manually turn off these features to regain space.

Key Points

• Source of the incident: Exclusive report by The Verge + 113 hot posts by Hacker News
• Scope of impact: All desktop users with Chrome AI enabled
• Core problem: 4GB weighted files are downloaded silently in the background, lacking user informed consent
• Solution: Turn off the AI function in Chrome settings to automatically release it

Background: What is Gemini Nano?

Gemini Nano is Google's lightweight AI model designed for end-side devices. Unlike Gemini Pro/Ultra running in the cloud, it runs inference directly on the local device. Chrome browser integrates Gemini Nano for the following features:

• Fraud Detection: Real-time analysis of suspicious content on web pages
• Writing Assistant: helps users write and optimize text
• Autofill: intelligently fill in form information
• Suggestions feature: contextual recommendations within the browser

Because the model weights file requires approximately 4GB of storage, Google automatically downloads it as part of the Chrome AI feature suite.

Key Impact

Adaptation suggestions

For different scenarios, you can adopt different strategies:

1. If storage space is tight: Go to Chrome Settings > System > Turn off the "Enhanced AI Protection" feature, and the 4GB files will be automatically removed during the next cleanup.
2. If you value the AI ​​function: Leave it on and understand that local AI protects privacy more than cloud AI.
3. If you are a developer: Check the OnDeviceModelExecution folder in the Chrome data directory to understand the specific location and occupation of the weight file.

Storage space check steps

• Open Chrome and enter chrome://version/ to view the profile path
• Enter the OnDeviceModelExecution/GeminiLite/ directory under this path
• Confirm whether the weights.bin file exists and its size
• Turn off unwanted AI features in Chrome settings

Deeper thinking: the hidden costs of on-device AI

This incident revealed a typical contradiction in the popularization process of end-side AI: users want intelligent functions, but are unwilling to pay the cost of storage and performance.

Google officials said, "The size of Gemini Nano will change with browser updates, and the model will be automatically uninstalled when the device resources are insufficient." But the problem is that these details only appear in the description of the developer documentation, rather than the user prompt interface when enabling the function.

This is in contrast to Apple Intelligence's approach in iOS 18 - Apple explicitly informs users of the storage requirements for local AI models when the device is first set up.

Connect to our tool ecosystem

If you are using Chrome to build a workflow with AI tools, it is important for you to understand how the local AI model works. The same local reasoning logic also applies to other local AI tools such as LM Studio and Ollama. The only difference is that Chrome is integrated and automatically run in the browser.

Google’s Gemini Nano positioning makes you think: when should you use local AI and when should you rely on cloud APIs?

Reference sources

• The Verge
• Hacker News
• Google ：Gemini Nano in Chrome

Internal link guidance

• Want to learn how to run local AI on your own computer? Watch: How to run local AI models on M4 Mac with LM Studio: A complete 30-minute tutorial
• Real case: Independent developers use AI tools to build a money-making system: 18-Year-Old Built a $5,000/mo SaaS With AI Agents — Zero Hand-Written Code