On-device AI is having its moment, and every phone maker wants credit for it. The pitch is that the intelligence lives in your hand, private and yours. But there is a question the marketing carefully skips: whose model is actually running? Owning the chassis is not the same as owning the brain inside it.

Apple is the clean case. It controls the silicon, the OS, and the model, top to bottom. When an Apple device runs something locally, Apple owns every layer of that stack. That vertical control is exactly why its privacy claims are legible: there is no one else in the room.

The Android side is murkier. When a handset maker announces on-device intelligence, ask where the capability originates. Often the model, the runtime, and the plumbing come from the platform layer, and the manufacturer supplies the hardware and a skin. In that arrangement the phone maker is not doing on-device AI so much as reselling a platform’s capability with its own badge on the box. Real ownership means you could swap the platform and keep the intelligence. Most cannot.

The question that separates owner from reseller

• Who trained the model, and who can update or revoke it?
• If the platform partnership ended tomorrow, does the on-device capability survive?
• Where does the inference actually run, and where does the telemetry actually go?

This is not a branding quibble. Ownership decides leverage. Whoever owns the model owns the privacy guarantees, the competitive moat, and the ability to change the terms later. A hardware badge on someone else’s intelligence is a lease dressed up as a deed. In a market this consequential, it is worth knowing which one you are actually holding.