Apple rented its way into AI: the Gemini/Siri economics

Apple spent two years looking like it was losing the AI race. It may simply have been running a different one.

This week, at its developer conference, Apple took the stage and unveiled a rebuilt Siri AI to the usual applause. Smarter, more conversational, and finally closer to the assistant it promised back in 2024. What it did not dwell on, from that stage, was whose intelligence was actually doing the work.

Apple reportedly pays Google around a billion dollars a year to run a custom Gemini model behind the new Siri. A billion sounds enormous until you hold it against the bonfire everyone else lit. OpenAI, Google, Meta and xAI have poured tens of billions into compute and training to control frontier models. Apple watched, waited, and rented one.

Then there is a small detail that should put smiles on the faces in Cupertino. Google already pays Apple around twenty billion dollars a year to remain the default search option on Apple devices. Apple takes twenty, hands one back, and keeps the change.

The lesson is not that Apple is frugal. It is that building the model was never necessarily the point. Apple controls the device, the operating system and the customer relationship. Intelligence is an input, and inputs can be rented.

While the industry argued over whose model was the cleverest, Apple treated the model as infrastructure and distribution as its moat. Then it stood on its own stage and kept the applause for itself. That is the part worth considering if you run a business and you are weighing how much AI to build yourself. The richest company on earth, with every means to train its own model, looked at the maths and chose to rent.

The expensive race was to build the model. The profitable race may be to own the moment the customer uses it.

If the richest company on earth is renting intelligence, why are you so sure you need to build it?