Apple brings Google’s Gemini models into its AI stack.
Problem – Scaling AI while keeping strict privacy guarantees Apple needs powerful language and vision models for Siri and Apple Intelligence, but its brand relies on on‑device privacy. The native Foundation models are deeply integrated into the platforms. At the same time, raw user data may not leave the device.
Solution path – Gemini as a remote‑inference service The Gemini model from Google is a central part of the architecture.
What worked
- Since June 8 2026 developers can embed the APIs in test apps; early feedback was positive because Gemini‑grade quality is available immediately.
What didn’t work
- Licensing Gemini costs Apple about $1 billion per year.
Trade‑offs
- Cost – High annual fee, but instant AI performance.
- Privacy – The design of Apple Intelligence is specifically designed to protect user privacy.
In short: The design delivers quick user value and requires ongoing budgeting.
Sources
- Apple Reveals New AI Architecture Built Around Google Gemini Models
- Apple Intelligence bringt leistungsstarke KI-Fähigkeiten für alltägliche Erlebnisse - Apple (DE)
- Siri 2.0: Apple präsentiert KI-Assistent mit Google-Gemini
- Joint statement from Google and Apple
- Apple reveals new AI architecture built around Google Gemini models
- Apple Plans to Use 1.2 Trillion Parameter Google Gemini Model to Power New Siri - Bloomberg