Apple Is Paying Google $1B Per Year to Power Siri With Gemini — What Changes for Developers

Apple is paying Google approximately $1 billion per year for access to Google's 1.2 trillion parameter Gemini model to power the next generation of Siri. The deal was announced January 12, 2026. The upgraded Siri launches at WWDC 2026 in June and replaces Apple's current 150 billion parameter Apple Intelligence models for complex reasoning tasks.

This is the largest AI infrastructure licensing agreement in the industry. Google already receives approximately $20 billion per year from Apple to be the default search engine in Safari. Now the flow reverses: Apple pays Google for AI. Both companies run on each other's money.

Why Apple Bought Instead of Built

Apple's internal AI development has consistently lagged. Apple Intelligence launched with iOS 18.1 in late 2024 using a 150B parameter model. It underperformed on complex reasoning, multi-turn conversation, and agentic tasks compared to GPT-4o and Gemini 2.0. User satisfaction with Siri was at a historic low and becoming a competitive liability against Samsung Galaxy AI and Google Pixel AI features.

Apple tried to fix this internally: it acquired several AI companies, hired away researchers from DeepMind and OpenAI, and built foundation models. None of it closed the gap fast enough. By late 2025, the internal models were still outclassed, and the product timeline for a competitive Siri was two or more years away.

The Gemini deal is a pragmatic admission that buying access to the world's most capable model beats a two-year development sprint. At $1B/year, it's expensive. But Apple's revenue miss from poor AI performance — reflected in slower iPhone upgrade cycles and lost services revenue — was costing more.

The Technical Architecture

Apple is not simply calling Google's API. The arrangement is more sophisticated:

On-device and Private Cloud Compute stays. Apple's on-device infrastructure continues handling tasks that can be processed locally — basic Siri commands, Apple Intelligence writing tools, on-device summarisation. This layer never sends data to external servers.

Gemini handles complex cloud tasks. When a query requires reasoning depth, knowledge, or contextual understanding that Apple's on-device models can't handle, it routes to Gemini running in Google's data centres. Apple has negotiated contractual protections preventing Google from using these queries for training.

Apple can edit Gemini in Google's infrastructure. The joint statement confirmed Apple can "directly reach and edit Gemini in its own data centres" — fine-tuning access, not just API access. Apple can customise Gemini's behaviour for Siri's task distribution and enforce Apple's privacy constraints at the model level.

The model size: Google's 1.2 trillion parameter model — the largest publicly acknowledged version of Gemini. Not Gemini Flash, not Gemini Pro. The full-scale infrastructure model. Apple's previous 150B model is being replaced by something 8x larger.

What the New Siri Actually Does

The redesigned Siri has four capabilities the current version cannot do:

Cross-app actions. Current Siri sets timers and sends messages. New Siri reads a restaurant review, checks your calendar, finds a free slot, and makes a reservation — across three apps in a single request.

On-screen awareness. Siri reads and understands what's currently displayed on your iPhone or Mac and acts on it. "What's the address in this email?" "Summarise this PDF." "Book a flight for the date in this message."

Multi-turn conversation. Maintains context across a session. You can follow up, correct, and refine without starting from scratch each time.

Personal context. With permission, Siri accesses emails, messages, calendar, photos, and health data to give contextually relevant responses. This is the capability that most requires Gemini's reasoning depth — it's why 150B wasn't sufficient.

What This Means for iOS Developers

Three direct implications:

SiriKit integrations don't need rewrites. The Gemini upgrade is in the model layer, not the API layer. Your existing Siri shortcuts, App Intents, and SiriKit domains continue working and get better responses automatically.

App Intents become genuinely useful. Apple has pushed App Intents (iOS 16+) as the mechanism for exposing app functionality to Siri. With a Gemini backbone, Siri's ability to chain App Intents across apps becomes actually reliable — "use my banking app to pay the person who just messaged me" stops being a demo and starts working. Developers who have implemented App Intents will see their integrations improve without code changes.

Enterprise apps need to evaluate the privacy architecture. Apps handling sensitive data — health, finance, legal — need to understand where processing happens. Simple commands: on-device, never leaves the phone. Complex reasoning: routes to Gemini via Apple's Private Cloud Compute, with Apple's contractual protections in place. Enterprise developers should verify whether their Siri integration use cases fall in the on-device or cloud category before enabling Siri for sensitive workflows.

Competitive Reshaping

Google is now inside 1.4 billion iPhones. Google pays Apple for search reach. Now Apple pays Google for AI. Google's model will generate responses inside every modern iPhone — the largest AI distribution channel in consumer tech history.

ChatGPT's Siri handoff role shrinks. Apple's iOS 18 deal gave OpenAI's ChatGPT access to Siri as an optional handoff when Siri couldn't answer. That integration remains in Settings, but Gemini now handles the complex queries that would have gone to ChatGPT. OpenAI's passive distribution advantage through Apple is significantly reduced. The ChatGPT vs Claude comparison remains relevant for direct API use — but on consumer iOS, Gemini wins.

Anthropic has no comparable deal. Claude is absent from this arrangement. Anthropic's distribution is API-driven and enterprise-focused — it has no passive reach inside iOS comparable to Google's new position.

Microsoft Copilot has no Apple deal either. Copilot is strong in enterprise and Windows. The Apple-Google arrangement further concentrates AI assistant distribution on Google across Android and iOS, leaving Microsoft competing primarily in the enterprise segment.

The Antitrust Watch

The DOJ is already investigating Google's search payments to Apple. A second billion-dollar payment — this time in the opposite direction, for AI — adds another dimension. EU regulators under the Digital Markets Act are also expected to scrutinise whether Google's dual position as Apple's search and AI infrastructure provider forecloses competition.

The deal was announced January 12. Regulatory proceedings, if they come, will likely follow the WWDC launch. Both companies appear to be moving fast to establish facts on the ground before antitrust challenges can take shape.

Key Takeaways

• Apple pays Google $1B/year for the 1.2 trillion parameter Gemini model — 8x larger than Apple's previous 150B model
• New Siri launches WWDC 2026 with cross-app actions, on-screen awareness, multi-turn conversation, and personal context access
• Architecture: on-device for simple tasks, Gemini cloud for complex reasoning; Apple contractually prevents Google from training on user queries
• For iOS developers: no rewrites needed; App Intents become more powerful automatically; enterprise apps should map which Siri use cases route to Gemini
• ChatGPT's Siri handoff role is reduced — Gemini handles the complex queries that previously went to OpenAI
• Google now generates responses inside 1.4 billion iPhones — the largest AI distribution channel ever
• DOJ and EU antitrust scrutiny incoming — the dual financial relationship between Apple and Google is attracting regulatory attention