The loud AI story is still about the cloud — bigger models, bigger data centres, bigger bills. The quieter story, the one that will matter more to most people, is that a surprising amount of it now runs on the device in your hand, with nothing leaving it.
Small, local, invisible
Transcription, translation, photo cleanup, summarisation, smart search across your own files: these increasingly happen locally. No round trip, no upload, no subscription meter running. It works on a plane. It works with the radios off.
That last part is the point. On-device inference is not just faster and cheaper; it is private by construction. The data that never leaves cannot be logged, leaked, or repurposed. For a whole category of everyday tasks, that is the feature — even if no billboard says so.
The AI that never phones home is the AI you can actually trust with the small stuff.
Nobody is marketing it hard yet because it is difficult to demo and impossible to upsell. But it is the version of this technology that fades into the tools you already use, which is usually where the real ones end up.