Apple Built Its New AI Foundation Models With Google

Apple did not build its new AI alone. At WWDC 2026, the company confirmed it partnered with Google, using the technologies behind Gemini, to create the next generation of Apple Foundation Models.

These models to run on-device and on Apple’s Private Cloud Compute servers, meaning the processing happens either on your device or on servers that Apple says do not store or share your data.

The result is a multimodal system that handles speech, text, and images all at once. It powers everything from Siri AI to image generation, writing tools, and visual intelligence across every Apple platform.

For the most capable Apple Silicon devices, Apple also built a second, more powerful on-device model. This one handles expressive Siri voices, advanced dictation, and more complex on-device reasoning.

Apple was clear on privacy. Unlike most AI providers that retain your interactions by default, Apple Intelligence uses on-device processing and Private Cloud Compute. Your data is only used to execute your request, and outside experts can verify this at any time.

Apple Intelligence second-generation architecture:
Co-developed with Google using Gemini technology
Adapted to run on-device and on Private Cloud Compute servers
Multimodal: understands speech, text, and images simultaneously
Photorealistic image understanding and generation
Second more powerful on-device model for capable Apple Silicon devices
Expressive speech generation for Siri voices
Advanced system-wide dictation
Personal context understanding via Spotlight semantic index
App actions across installed apps
On-screen awareness and real-time assistance
Broad world knowledge via web, processed on Private Cloud Compute
Privacy: no data stored, no access by Apple, external verification available
Free with iOS 27, iPadOS 27, macOS Golden Gate, watchOS 27, visionOS 27