Chrome downloads multi‑GB AI models

Chrome is turning into something a little different. Not just a browser that fetches web pages, but a host for local AI models that live on your machine. That shift is why some users are suddenly finding multi-gigabyte Gemini Nano files inside Chrome’s internal folders. The surprise is not that Google wants AI in Chrome — that part has been obvious for months. The surprise is that the browser can manage those models in the background, with storage and bandwidth costs that many people would not expect. ### What is Chrome actually downloading? The thing showing up on disk is Gemini Nano, Google’s local language model for Chrome’s built-in AI features. Chrome’s developer docs say the browser manages these models itself, and that downloads, updates, and swaps happen automatically once a supported AI feature needs them. Xataka traced one example to a folder called `OptGuideOnDeviceModel`, where a Mac showed 4.27 GB in use. ### Why does it happen in the background? Basically, Google designed this to feel invisible. Chrome says model management is “seamless,” and the initial download can continue even if the tab closes. If the connection drops, the download resumes later. If the browser closes, Chrome can resume on restart within 30 days. That is convenient for developers — more like an app platform than a simple web client. ### When does the download start? The clean version is: when an AI feature calls for it. Chrome says the first call to a built-in AI API’s `create` function can trigger the initial Gemini Nano download. But there is a catch. Google also says that, in some cases, even an `availability` check can trigger a download shortly after a fresh profile start. How Chrome prepared the feature is blurrier than most people would assume. ### How big can this get? Big enough that Google tells developers to expect serious headroom. Chrome’s built-in AI docs say supported systems need at least 22 GB of free space on the volume containing the Chrome profile, plus either a capable GPU or a fairly strong CPU-and-RAM setup. Google also says the exact model size can vary with updates. That helps explain why the recommended space is much higher — the browser wants room for downloads, updates, and different model variants. ### Is this tied to Gemini in Chrome? Partly, but not only that. Google’s big Chrome AI push in late 2025 included Gemini in Chrome, AI Mode in the address bar, scam detection, and more built-in AI features. At the same time, Chrome expanded Gemini Nano support beyond GPUs to CPUs in Chrome 140, which means many more desktops and laptops can now run these local models. More compatible devices means more chances for these downloads to appear. ### Is local AI better for privacy? In one narrow sense, yes. Google says no data is sent to Google or third parties when the model runs locally for these built-in AI APIs. That is the upside of on-device inference — lower latency and less need to ship prompts to a server. But privacy is not the only issue here. Users also care about consent, disk usage, and control — which are not the same thing as transparent behavior. ### Can users check or remove it? Chrome says installed model versions can be checked manually at `chrome://on-device-internals`, and its model-management docs include deletion as part of the lifecycle. Google also has a help page stating that Chrome may download on-device generative AI models in the background so supported features stay ready. So this is not a bug in the usual sense; it's about users discovering that design after the files are already there. ### Bottom line The real story is not one weird 4 GB folder. It is that Chrome now ships with an AI layer that can fetch, update, and maintain local models on its own. That brings speed and some privacy benefits, but it also changes the bargain people thought they had with their browser.