No. AI that runs on your phone doesn’t need 5G or any connection once the model is on the device. Cloud chatbots like ChatGPT and Gemini need a connection for every message, and 5G only shortens the trip. Most of the wait for a cloud reply is the server generating text, not the network, so upgrading from good 4G to 5G rarely changes much.
That’s the practical core of all the “AI plus 5G plus edge computing” talk. Here’s the rest.
How much of a cloud AI reply is the network? #
Less than you’d think. A typed question is a few kilobytes. Your phone sends it, a data center runs the model, and the answer streams back word by word. The model generating tokens is most of the wait, and on a busy day you may also queue behind other users.
Where a fast connection does help:
- Uploading photos, voice or video for a cloud model to analyze.
- Live voice conversations with a cloud assistant, where every pause shows.
- Weak or congested signal, where requests time out or stall halfway through an answer.
Where it doesn’t help: the thinking time of a big model, or anything when you have no signal at all.
What is edge computing, and is your phone “the edge”? #
Edge computing means doing the work close to where the data is created instead of in a distant data center. The “edge” can be:
- Your phone, watch or laptop. This is the edge most people actually use. When a model runs on your phone, it’s edge AI.
- Servers near the cell tower, which carriers call multi-access edge computing. They cut network distance for things like cloud gaming, but you rarely know if an app uses them.
When telecom companies talk about “AI, 5G and edge converging,” they mostly mean the second kind, aimed at factories, cars and their own networks. For a person with a phone, the meaningful edge is the device in your hand. We cover more everyday examples in what is edge AI.
When do you actually need a connection for AI? #
| Task | Cloud chatbot | Model running on your phone |
|---|---|---|
| Chat, rewrite, brainstorm | Needs connection | Works offline |
| Ask about a photo | Needs connection | Works offline with a vision model |
| Ask about a PDF you have | Needs connection | Works offline in apps that support documents |
| Current news, prices, scores | Yes, with web search | No. The model only knows its training data |
| Generate images or video | Yes | Mostly not on phones today |
| Very long, hard reasoning | Better | Possible but slower and less accurate |
A local model knows nothing after its training cutoff and can’t browse. Everything else in the top half of that table works in airplane mode.
Which phone AI features work offline today? #
- Local AI chat apps. Personal LLM downloads an open model such as Qwen 3.5 or Gemma 4 once, then chats, reads photos and answers questions about attached PDFs with no connection. Switch on airplane mode and it keeps answering.
- Keyboard dictation on many iPhones and in many languages, which Apple processes on-device.
- Google Translate, including camera translation, once you download the language packs.
- Some Apple Intelligence features. Apple uses a mix of on-device processing and its Private Cloud Compute servers, so some tasks work offline and others don’t.
Does bad signal make on-device AI better? #
It makes it more useful, not faster. On a train, in a basement, or on a plane, a cloud chatbot either stalls or fails. A local model runs at the same speed whether you have five bars or none, because the network isn’t involved.
It also means no mobile data use. Once the model file is downloaded (do that on Wi-Fi, since models are 0.8 to 6 GB), chatting uses none. The one exception in Personal LLM is ads: the free app shows ads from Google AdMob when you’re online, and none load offline.
Is 5G greener or better for privacy? #
Neither, in any way you’d notice. A faster link doesn’t change what’s sent. With a cloud chatbot your text still goes to the provider’s servers over 5G, 4G or Wi-Fi alike. Privacy depends on where the model runs, not how fast the pipe is. The carbon side is small either way, as we worked out in the carbon footprint of cloud vs on-device AI.
So should you care about 5G for AI? #
Care about it if you use cloud voice assistants a lot, upload many photos to cloud models, or live somewhere 4G is congested. Otherwise your existing connection is fine, and the bigger question is which tasks you’d rather keep on the phone entirely. For that, start with what on-device AI is and why it matters for privacy.
Frequently asked questions #
Does ChatGPT work without internet? #
No. ChatGPT runs on OpenAI’s servers, so every message needs a connection. To chat with AI offline you need an app that runs a model on your phone, and you have to download the model once while connected.
Do Apple Intelligence or Gemini need a connection? #
Partly. Apple Intelligence runs some features on the iPhone and sends others to Apple’s Private Cloud Compute servers. The Gemini app mainly runs on Google’s servers, though some Android phones run a smaller on-device model for specific features. Test a feature in airplane mode to see which kind it is.
What AI works in airplane mode? #
Anything whose model lives on the phone: local AI chat apps, on-device dictation, offline translation packs and on-device photo tools. The first setup or model download needs a connection, then they run offline.
Does on-device AI use mobile data? #
Not while it’s answering. The only data use is the initial model download, which can be several gigabytes, so do it on Wi-Fi. Some free apps also load ads when you’re online.