What you need to know
- AICore can temporarily use large storage when updating AI models in the background on your device.
- Google keeps both old and new AI models for up to three days as a fail-safe during updates.
- Storage used by AICore is automatically freed once the new AI model is confirmed stable.
If you have the AICore app installed on your Android phone and have noticed it taking up an unusually large amount of storage, Google has finally explained why that happens.
Over the past few years, companies like Google and Samsung have been pushing on-device AI across their phones. On most flagship Android phones, a big part of that functionality is handled by the AICore app.
The AICore app is essentially a system service that powers on-device Gemini Nano on Android. It enables offline and private AI features like smart replies and notification summaries on devices such as the Pixel 10 Pro. But if you've noticed it using a lot of storage, you're not alone — some users have reported it taking up as much as 11GB.
Article continues belowGoogle has now updated its AICore support page to explain this behavior. The company says you may "occasionally find that this service uses a larger amount of storage than expected." As per the support page, this happens when it's updating to a new version of an AI model in the background.
To make sure everything works reliably, AICore temporarily keeps both the old and new versions of the AI model on your device for up to three days. This acts as a fail-safe, allowing the system to roll back instantly if something goes wrong during the update, without requiring you to download large files again.
I get why Google is doing this, and it's a smart fail-safe. But 10GB+ for a background service is still wild. At the very least, there should be a way to limit or manage this manually.
Google also says that this extra storage is freed up automatically once the update is confirmed to be stable, so you don't need to take any action.
Honestly, this approach does make sense. If the app removed the older model right away and the update failed, you'd have to re-download several gigabytes of data and lose access to AI features in the meantime. This method makes the whole process more reliable.
