Apple iOS 27 Siri AI — Google Gemini partnership

Apple Just Handed Google Control of Siri. Here's What That Actually Means.

iOS 27 shipped September 14 with Siri AI in public beta, built on Google Gemini. It's the first non-Apple foundation model to power a shipping Siri. Daily caps apply. The EU is excluded on iPhone. And the data boundary — where your queries leave your device — is not publicly disclosed.

TL;DR

- iOS 27 shipped September 14 with Siri AI in public beta — rebuilt from the ground up using Google Gemini models. This is the first time a non-Apple foundation model has powered a shipping version of Siri. - Access is not automatic. Users opt in via Settings, may join a waitlist, and need iPhone 15 Pro or newer. English only for now; French, Japanese, Korean, Portuguese, and Spanish coming in October. - Siri AI is not available in the EU on iPhone or iPad due to the Digital Markets Act. EU users can access it on Mac and Vision Pro only. - Daily usage caps apply. Apple has confirmed expanded limits will "be available for a fee in the future." - This is John Ternus's first major software launch as Apple CEO. He succeeded Tim Cook on September 1. His opening statement as CEO was about the new Siri. - The iPhone 18 lineup — including the iPhone Duo, Apple's first foldable iPhone — launches this week.

Apple spent two years explaining why it could not ship a smarter Siri. On September 14, it shipped one. It is built on Google Gemini.

That sentence takes a moment to settle. Apple has built its competitive moat on in-house hardware and software integration for decades. The M-series chips. The Apple Silicon transition. The argument that Apple's unique value is the seamless integration of components it controls end to end. The new Siri is built on a model Apple does not train, does not fully control, and did not invent.

This is not a minor partnership. It is a structural decision about where frontier AI capability comes from and what Apple is willing to trade to get it.

What Siri AI actually does

The old Siri was a command dispatcher. You said a thing; it either parsed the thing into an action or it did not. The failure mode was familiar: Siri confidently doing nothing, or confidently doing the wrong thing.

The new Siri is context-aware. It can read your messages and emails before you ask a question, so you can say "when did he say he was arriving?" without specifying who "he" is. It can look at your calendar, recent photos, and current screen and make connections the old Siri required multiple explicit prompts to approximate. It can complete multi-step tasks across apps that would have stalled the previous version mid-command.

Apple CEO John Ternus described the approach at the iPhone 18 launch event: "It's personal intelligence that's actually personal." The word "personal" appearing twice in one sentence is deliberate. The new Siri's differentiation is not raw capability — it is access to context that other assistants do not have by default. Your messages. Your calendar. What is on your screen right now.

Processing splits between on-device computation and Apple's Private Cloud Compute infrastructure. For queries that require more capability than the on-device models can handle, the request goes to cloud servers. Apple has not publicly specified the exact boundary between on-device and cloud processing — meaning the line where your personal data leaves your device and reaches Google's infrastructure is not disclosed.

The Google question

Google confirmed the partnership in April 2026, describing itself as Apple's "preferred cloud provider" for developing next-generation Apple Foundation Models. Apple applied a distillation process to Gemini models, producing smaller derivatives tuned for Apple Silicon.

The arrangement means Google has a new distribution channel at a scale no other AI partnership has yet achieved. The iPhone installed base is over a billion active devices. Every Siri AI query that escalates to cloud processing goes through infrastructure where Google is the named partner.

For Google, this is search economics applied to AI: reach the user at the point of query, regardless of whose device or whose app they are using. For Apple, it is a pragmatic acknowledgment that training frontier models at Gemini-grade capability is not currently within Apple's published research output — and that shipping a competitive assistant required partnering with someone who could.

Apple dropped its earlier exclusive arrangement with ChatGPT, introduced with iOS 18. The new platform is explicitly open: any AI service can integrate via the App Store, and users can set a preferred model for specific tasks. OpenAI's ChatGPT integration remains available. Google Gemini is the default and the foundation.

The EU gap and the daily cap

Two things Apple has not resolved cleanly.

Siri AI is not available on iPhone or iPad in the European Union at launch. Apple cited the Digital Markets Act — specifically the obligations that would apply to a deeply integrated AI assistant with access to personal data across all Apple apps — as the reason. EU users can access Siri AI on Mac and Vision Pro. Apple has not committed to a timeline for iPhone and iPad availability in the EU.

The EU has more than 100 million iPhone users. "Not available" for this release is a significant gap. Whether it reflects genuine regulatory uncertainty or leverage in ongoing DMA negotiations is a question Apple's legal team could answer more precisely than its product team has.

The daily usage caps are a different kind of signal. Apple has confirmed that Siri AI features relying on cloud servers will have daily limits, and that expanded limits will be available for a fee. The exact structure of the paid tier has not been disclosed. What this establishes is that the new Siri is, at some usage threshold, a subscription product — or at minimum a product with a consumption model attached to it.

That is new for Apple software. iCloud storage and Apple TV+ are services. iOS features have historically been either free or included in hardware price. A Siri with a daily cap that expands for a fee is a different business model from anything Apple has shipped at the operating system level before.

What this means for enterprise AI strategy

Most of the coverage of iOS 27 is treating Siri AI as a consumer story. It is also an enterprise architecture question.

Apple devices are the dominant endpoint in many enterprise environments. A Siri that can read emails, access calendars, and understand on-screen context — and that escalates queries to cloud infrastructure with an undisclosed data boundary — changes what enterprise mobile device management policies need to address.

The data boundary question is not hypothetical. When a user asks Siri AI to summarize an email thread using their iPhone's on-device capability, the data does not leave the device. When the same query requires cloud processing, it does. Apple's Private Cloud Compute architecture is designed to process cloud queries without retaining data on Apple's servers. What it does not specify is the path that data takes through Google's infrastructure during the processing of Gemini-powered responses.

Enterprise mobility management teams that have built policies around "data stays on the device" assumptions for Apple devices will need to revisit those assumptions for Siri AI cloud-processed queries.

The daily cap also matters for enterprise deployments. If Siri AI usage in a high-volume enterprise environment hits daily limits, the assistant reverts to the previous Siri as a fallback. Workflows designed around Siri AI capability should account for fallback behavior.

The Ternus factor

iOS 27 is the first major product launch under John Ternus, who became Apple CEO on September 1, succeeding Tim Cook. Ternus built his reputation at Apple running hardware engineering — he was responsible for the M-series silicon transition and the product engineering that made Apple's in-house chip strategy work.

His opening statement as CEO was about Siri AI. "It's personal intelligence that's actually personal."

That framing — Ternus, the hardware engineer, defining his tenure through an AI assistant built on a partner's foundation model — is the signal worth reading. Apple's next chapter on AI is not going to be about training frontier models in-house. It is going to be about what Apple can do with frontier models it partners to access, on hardware it controls, with a data boundary it can credibly claim is private.

Whether that strategy wins in the AI era is the question Ternus's tenure will answer.