Apple Taps Google Gemini for New On-Device AI Architecture

Apple just admitted it can't do state-of-the-art AI alone — it's co-developing foundation models with Google using Gemini technology. The two companies described the collaboration as "deep," and Apple says it unlocks a "huge upgrade" for Apple Intelligence with multimodal support including image understanding and generation.

Foundation Models Co-Developed, Not Just Licensed

These are not off-the-shelf Gemini models dropped into iOS. Apple says its Foundation Models are adapted to run both on-device and on servers through its existing Private Cloud Compute infrastructure. A higher-power version adds speech generation, improved dictation accuracy, and stronger natural language understanding — though Apple hasn't said which devices qualify for that tier.

The new architecture brings realistic image creation, advanced photo editing, and visual question answering. That's a sharp departure from the text-only, summarization-and-rewrite approach of earlier Apple Intelligence releases.

A System Orchestrator, Not Just a Model

At the center sits a system orchestrator that coordinates Apple Intelligence features across all of Apple's platforms. It tailors responses based on the active app and the user's current task — Apple calls it "truly system-wide intelligence." The orchestrator decides where to run inference: on the neural engine, GPU, or Private Cloud Compute servers, depending on the request's complexity.

Privacy guarantees remain unchanged: user data is only used to execute the immediate request and is not accessible to Apple or third parties. Apple says outside experts can verify those guarantees "at any time."

Contrast Play Against Competitors

Apple framed the announcement as a contrast to competitors it characterized as "racing forward" without regard for users. That's a direct shot at companies shipping cloud-dependent AI with weaker privacy — and a signal that Apple sees co-opting Google's model technology as the only way to catch up without renting user data.

Whether third-party auditors actually get access to the Private Cloud Compute infrastructure at scale remains to be seen. But the technical architecture — foundation models co-developed with Google, a privacy-preserving orchestrator, on-device fallback — gives Apple a credible story for enterprise and security-conscious users looking for state-of-the-art generative AI without surrendering data.

Source: Apple reveals new AI architecture built around Google Gemini models
Domain: macrumors.com