Google is involved, and Apple has described that involvement in exactly one published sentence. Apple’s June press release says its next-generation Apple Foundation Models “were custom-built in collaboration with Google and its Gemini models for deeply integrated Apple Intelligence experiences,” and nothing on Apple’s consumer pages goes further than that.
Everything else circulating about the arrangement, including the widely repeated claim that a 1.2-trillion-parameter custom Gemini model runs on Apple’s servers, is a reported figure rather than an Apple specification. Here is the confirmed wording, where Siri AI requests actually run, and how to separate the two.
Key facts: what Apple has confirmed
| Item | What Apple says | Source |
|---|---|---|
| Google collaboration | Apple Foundation Models “were custom-built in collaboration with Google and its Gemini models for deeply integrated Apple Intelligence experiences” | Apple’s press release, June 8 |
| Where models run | On-device, and on Apple servers through Private Cloud Compute | Apple’s Apple Intelligence page |
| Third-party branding | No Gemini or other outside model is named anywhere on Apple’s Apple Intelligence, Siri or iOS 27 pages | Apple’s iOS 27 page |
| Orchestration | A “System Orchestrator” that draws on the on-device Spotlight index and an “App Toolbox” | Apple’s Apple Intelligence page |
| Cloud model name | “AFM 3 Cloud models” in Shortcuts | Apple’s Apple Intelligence page footnotes |
| Model size | Apple publishes no parameter count for any of its models | Apple’s Apple Intelligence page |
| Privacy claim | Personal data “is not stored nor made accessible to Apple or anyone else” | Apple’s Apple Intelligence page |
| Developer tools | Xcode 27 agentic coding works with Anthropic, Google and OpenAI models | Apple’s press release, June 8 |
The one sentence Apple published
Read the confirmed line closely and it says less than it is often taken to say. Apple’s wording is that its own Foundation Models were built in collaboration with Google and its Gemini models. It does not say that Siri AI sends your questions to Google, that Google operates any part of the service, or that a Gemini model answers a request on your iPhone.
Don’t miss the best of The Mac Observer
Set us as a preferred source and our Apple reporting ranks higher in your Google Search results and Discover feed — one tap, no account changes.
Apple also has not said what the collaboration consisted of, how long it ran, whether money changed hands, or which of the two companies trained what. The sentence appeared once, in the June press release that introduced Siri AI and the rest of iOS 27, and Apple has not expanded on it since, including in the September 9 announcements at the “Surprise and shine” event.
Where Siri AI requests actually run
Apple’s architecture description is consistent across its pages: requests are handled by Apple Foundation Models running on the device, and heavier requests go to Apple servers through Private Cloud Compute. Apple’s Apple Intelligence page says a “System Orchestrator” decides what to use, drawing on the on-device Spotlight index and an “App Toolbox” for actions inside apps.
The privacy promise attached to that is specific. Apple says personal data “is not stored nor made accessible to Apple or anyone else,” and that “outside experts can continue to verify this privacy promise at any time.” On the hardware side, the A20 Pro chip in the iPhone 18 Pro carries a dual 16-core Neural Engine that Apple credits with double the AI processing of the A19 Pro, which is the on-device half of the split.
Server-side work has a cost attached. Apple’s pages say features that rely on server models are “subject to daily usage limits, including but not limited to Siri AI, Intelligent Photo Editing Tools, Image Playground, and AFM 3 Cloud models in Shortcuts,” that limits “may vary by feature, request complexity, system demand,” and that “expanded access to such features will be available for a fee in the future.” Apple’s June release adds that increased access comes with most iCloud+ plans.
Confirmed, reported and never stated
| Claim | Status |
|---|---|
| Apple Foundation Models were custom-built in collaboration with Google and its Gemini models | Confirmed by Apple’s press release |
| Siri AI runs on-device and on Private Cloud Compute | Confirmed by Apple’s Apple Intelligence page |
| Some features have daily limits and a paid tier is coming | Confirmed by Apple’s Apple Intelligence page |
| A custom 1.2-trillion-parameter Gemini model powers Private Cloud Compute | Reported; not confirmed by Apple |
| Siri AI passes queries to Google’s own servers | Not stated by Apple |
| Gemini is named or branded anywhere in iOS 27 | Not stated by Apple |
| ChatGPT or another chatbot is built into Siri AI | Not mentioned on any Apple page |
| Xcode 27 agentic coding uses Anthropic, Google and OpenAI models | Confirmed, but it is a developer tool, not part of iOS 27 |
The parameter count is worth singling out. It appears in third-party reporting published around the WWDC announcement, it is not on any Apple page, and Apple has never published a parameter count for its own models either. Treat it as an expectation and not a specification.
The one place Apple does name other companies’ models
Apple’s June developer-tools release says Xcode 27 adds agentic coding that works with models from Anthropic, Google and OpenAI. That is a Mac development tool for writing software, and it has no connection to what runs on a customer’s iPhone. It is also the only Apple text that names outside models in a product context, which is part of why the Foundation Models sentence drew attention.
What Apple has not said
- What the collaboration with Google involved, when it happened, or whether it is a commercial agreement.
- The parameter count, architecture or name of any Apple Foundation Model, on-device or in the cloud.
- Whether Gemini models run inside Private Cloud Compute, or only informed how Apple’s models were built.
- What the daily usage limits are in numbers, or what the future paid tier will cost.
- Whether the collaboration affects the regions where Siri AI is held back, including the EU and China.
Siri AI arrives as a beta with iOS 27 on Monday, September 14, free, in English first, with French, Japanese, Korean, Portuguese and Spanish following in October, according to Apple’s press release. If Apple publishes more about the Google collaboration we will update this article, and our event recap covers everything Apple confirmed on September 9.
