Apple’s chief software architect, Craig Federighi, revealed the company’s new AI framework at a quiet event after Tim Cook’s keynote. The design keeps data on the device whenever possible, then reaches out to cloud‑based models that live on third‑party servers. Federighi emphasized that Apple’s privacy guarantees remain intact even when the AI runs on Google’s infrastructure.

The system relies on a tiered stack of models. On‑device devices use AFM 3 Core, a Gemini‑based engine co‑created with Google, while newer models with 12GB+ RAM and the latest chips run AFM 3 Core Advanced, which taps local storage for richer dictation and a more expressive Siri voice. For complex queries, phones contact cloud models like AFM 3 Cloud, ADM 3 Cloud, and AFM 3 Cloud Pro.

To satisfy Apple’s privacy stance, the company introduced Private Cloud Compute for third‑party hardware, layering Nvidia’s Confidential Computing, Intel’s Trust Domain Extensions, and Google’s Titan chip. Apple also maintains a cryptographically verifiable ledger of Google Cloud devices and only accepts software signed by Apple. The move lets Apple keep its AI privacy promises while leveraging Google’s GPU power. This strategy positions Apple against competitors who rely solely on in‑house servers.