Apple Is Paying Google $1B/Year to Power Siri with Gemini — What Developers Need to Know

Apple had three choices for who would power the next generation of Siri: OpenAI, Anthropic, or Google. It chose Google, and it is paying approximately $1 billion per year for the privilege. The deal makes Google Gemini the AI brain of every iPhone, iPad, and Mac running iOS 26.4 — placing Google's language model inside the pocket of 2 billion Apple device users, running through Apple's privacy infrastructure.

This is not a minor API integration. It is a fundamental architectural decision that changes what Siri can do, who owns the intelligence layer of Apple's ecosystem, and what every iOS developer needs to understand about how AI features work on Apple platforms going forward.

Why Apple Chose Google Over OpenAI and Anthropic

Apple evaluated all three major AI providers before selecting Google. The official statement from Apple: "After careful evaluation, we determined that Google's technology provides the most capable foundation for Apple Foundation Models and we're excited about the innovative new experiences it will unlock for our users."

The technical reasoning, based on reporting from MacRumors and CNBC, breaks down across three dimensions.

Model capability: Gemini's 1.2 trillion parameter architecture and multimodal capability — handling text, images, audio, and video natively — matched what Apple needed for on-device AI that understands the full context of what's on screen. Apple wanted a model that could look at a document, understand its content, and take action on it — not just process text queries.

Privacy architecture compatibility: Apple's Private Cloud Compute (PCC) is the infrastructure that allows Apple to run cloud AI processing without Apple (or anyone else) having access to user data in plaintext. Google's infrastructure proved more compatible with PCC's requirements than OpenAI's. The actual inference runs through Apple's privacy layer — Google's model processes the request, but the request is structured in a way that prevents Google from retaining user data.

Commercial terms: OpenAI and Anthropic both wanted arrangements that would give them visibility into Apple's user interaction data — valuable training signal. Apple refused. Google accepted Apple's privacy requirements. The $1 billion per year Google receives is payment for capability without data access.

What iOS 26.4 Siri Can Actually Do

The Gemini-powered Siri arriving in iOS 26.4 — and expanded further in iOS 26.5 and iOS 27 — is not the same assistant that has been the butt of jokes for years. The new capabilities are specific and meaningful.

On-screen awareness: Siri can now see what is on your screen and understand it in context. If you are reading an email from a colleague asking you to schedule a meeting, you can say "Siri, schedule this" and it understands the email content, the proposed time, and creates the calendar event without you repeating any information. The model reads the screen; you give the intent.

Personal context: Siri builds and maintains a personal context model — who your frequent contacts are, what your recurring calendar patterns look like, what apps you use for which purposes. Requests are interpreted against this context. "Message the usual Thursday group" is understood because Siri knows your Thursday group from your message history.

Deep per-app control: Siri can now take actions inside third-party apps at a more granular level than before. The AppIntents framework, which developers use to expose their app's capabilities to Siri, is the technical backbone. Apps that have implemented AppIntents properly can receive complex instructions through Siri — not just "open the app" but "in Todoist, mark the project review task as complete and move the deadline to Friday."

Cross-app workflows: Siri can now chain actions across multiple apps in a single request. "Take the attachment from the last email from Sarah, convert it to a PDF, and share it in the team Slack channel" is a three-app workflow Siri can execute. This required both the Gemini model's reasoning capability and Apple's tight OS-level integration.

The Privacy Architecture: How Apple Keeps Google Out of Your Data

The obvious concern with this deal: Google, a company whose business model is built on personal data, now has access to Siri queries from 2 billion Apple devices. The concern is legitimate. Apple's answer is the Private Cloud Compute architecture.

When a Siri request requires cloud processing (on-device Apple Silicon can handle simpler queries locally), the request is processed through PCC. Inside PCC, the request is structured as follows: the user's query and relevant context are cryptographically isolated from any identifier that would allow Google to link the query to a specific user. Google's Gemini model processes the query as a stateless request — it sees the input, generates the output, and has no mechanism to store or associate that data with a user identity.

Apple has made PCC's architecture publicly verifiable — security researchers can inspect the system to confirm the privacy guarantees. This is a technical commitment, not just a policy commitment.

The practical reality: Google's Gemini improves its models through training on user interactions. The Apple deal structurally prevents Gemini from using Apple user interactions as training data. Google gets $1 billion per year and the reputational benefit of powering Apple Intelligence. It does not get the data.

What This Means for Developers

If you are building iOS apps, the Gemini-powered Siri changes your development calculus in several ways.

AppIntents implementation is now mandatory for AI-first apps. The new Siri capabilities are delivered through Apple's AppIntents framework. Apps that have not implemented AppIntents are invisible to the new Siri — it cannot control them, cannot chain them into workflows, cannot extract their data for context. As Siri becomes genuinely useful, users will expect their apps to be Siri-accessible. Apps that are not will feel second-class.

The on-screen awareness changes app design assumptions. When Siri can read your app's screen content and understand it, the visual hierarchy and content structure of your app becomes AI-accessible data. Apps that display information clearly and semantically will work better with Siri than apps with visually complex but semantically opaque interfaces.

Privacy-first design is now a competitive advantage. Apple's PCC architecture means Siri can be trusted with sensitive information — medical records, financial data, legal documents. Apps in sensitive categories (health, finance, legal) that integrate AppIntents can offer AI-assisted workflows that would be impossible on platforms without equivalent privacy infrastructure.

The developer API for Siri is expanding. Apple is releasing new AppIntents capabilities alongside iOS 26.4 that allow deeper integration points. Developers who adopt early will have Siri integration that competitors without AppIntents support cannot match.

Why This Deal Changes the Competitive Landscape

Apple chose Google over OpenAI — the company that arguably invented the consumer AI era — for a billion-dollar per year deal. This has implications beyond Apple's ecosystem.

For OpenAI, losing the Apple deal is significant. Apple devices are where a large percentage of ChatGPT's users access the product. A Gemini-powered Siri that handles common AI queries natively reduces the number of times iPhone users open the ChatGPT app. The substitution effect is real.

For Google, the deal is a defensive masterstroke. Google was facing a scenario where Apple's 2 billion devices became an OpenAI distribution channel. Instead, those devices are now a Gemini distribution channel. The $1 billion per year Google pays Apple is, in part, insurance against that alternative.

For Anthropic, not being chosen confirms that Claude's strength is enterprise and developer use cases, not consumer OS integration. Anthropic's constitutional AI approach and enterprise positioning did not fit Apple's requirement for a model that would run inside a consumer privacy architecture at scale.

Key Takeaways

• Apple is paying Google $1 billion per year to power Siri with Gemini — chosen over OpenAI and Anthropic based on model capability, privacy architecture compatibility, and Google's acceptance of Apple's no-data-access terms
• iOS 26.4 Siri capabilities: on-screen awareness (reads what's on your screen), personal context (knows your contacts, habits, patterns), deep per-app control via AppIntents, and cross-app workflow chaining
• Private Cloud Compute keeps Google blind to user identity — queries are cryptographically isolated, Gemini processes stateless requests, Google cannot use Apple user interactions as training data
• AppIntents implementation is now essential for iOS developers — apps without AppIntents are invisible to the new AI-capable Siri; the framework determines whether your app participates in the AI-assisted workflow era
• OpenAI lost the Apple deal because it wanted data access Apple refused; Google accepted Apple's privacy terms; the substitution effect on ChatGPT usage on iOS is real and will be visible in OpenAI's consumer metrics
• Google's strategic win: 2 billion Apple devices become Gemini distribution points instead of OpenAI distribution points — the $1B annual payment is partly insurance against an OpenAI-powered Apple ecosystem
• iOS 26.5 and iOS 27 continue the rollout — some features slipped from iOS 26.4 due to delays; full Gemini-Siri integration completes across 2026