Chrome reserves 4GB for AI
- Google Chrome is preloading Gemini Nano files on some desktop machines, using local disk so browser AI features and web AI APIs stay ready. - The key number is 4 GB — but Chrome’s own docs also say built-in AI needs 22 GB free first, then purges the model below 10 GB. - That matters because Chrome is treating local AI like a managed browser component now, not a one-off experiment users launch manually.
Chrome is starting to act like an AI runtime, not just a browser. That is the real story here. A bunch of people noticed a roughly 4 GB Gemini Nano payload sitting inside Chrome’s local files this week, and Google’s own docs make clear this is part of a broader system: Chrome downloads, updates, swaps, and purges on-device models in the background so AI features are ready when needed. (9to5google.com) ### What is Chrome actually storing? The storage is tied to Gemini Nano, Google’s small local language model for desktop Chrome. Chrome uses it for built-in AI APIs like summarizing, writing, rewriting, and prompting, and Google also says the same local model helps power browser features such as scam detection without sending data to the cloud. (developer.chrome.com) ### Why does the number look like 4 GB? Because that is the rough size people are seeing on disk for the downloaded model files — including a large `weights.bin` file in Chrome’s optimization-guide area. But Chrome’s developer docs frame 4 GB as the installed payload, not the eligibility threshold. To even use the Gemini Nano-backed APIs, Chrome sa(developer.chrome.com)hrome profile. (9to5google.com) ### So does Chrome download it for everyone? Not exactly. Chrome says the model is downloaded on demand, based on the user’s hardware, and only if the device clears a set of checks. Chrome benchmarks the machine, decides whether to use a larger or smaller Gemini Nano variant, and can even fall back to CPU inference if the system qualifies. If the hardware does not meet the requirements, Chrome skips the download. (developer.chrome.com) ### Why are people seeing it “silently” appear? Because Chrome manages the whole thing like a browser component. Downloads can continue in the background, resume after interruptions, and even restart after the browser reopens. Google’s model-management docs also say that sometimes a simple availability check can kick off a (developer.chrome.com)eel like nothing happened — until the disk space is gone. (developer.chrome.com) ### What if your laptop is low on space? Chrome says the model is disposable. If free storage drops below 10 GB after the download, Chrome removes the model from the device and can fetch it again later once the space requirements are met. That is an important detail — Chrome is not reserving 4 GB forever no matter what. It is more like a managed cache for a browser-level AI subsystem. (developer.chrome.com) ### Can you turn it off? Google says yes, though the controls are still uneven. The company told 9to5Google that a setting to turn off and remove the model started rolling out in February 2026, and its help pages describe Chrome downloading on-device generative AI models in the background for browser and web features. Some users can reach AI contro(developer.chrome.com)llout stage. (9to5google.com) ### Why is this a bigger deal than 4 GB? Because it shows where browsers are heading. Chrome is no longer treating AI as a tab that calls a server somewhere else. It is treating AI like spellcheck, Safe Browsing, or a media codec — a built-in capability with lifecycle management, hardware gating, background updates, and its own int(9to5google.com)g part of the browser’s plumbing. (developer.chrome.com) ### Bottom line? The surprising part is not that Chrome can use 4 GB for AI. The surprising part is that Google has already built the machinery to make local models feel normal — downloaded when useful, updated quietly, and deleted when space gets tight. That makes this look less like a quirky tweak and more like the default shape of browser AI from here on out. (developer.chrome.com)