Apple's partnership with Google to rebuild Siri represents a significant shift in the company's AI strategy, with reports suggesting the deal costs Apple approximately $1 billion per year. Announced on January 12, 2026, this multi-year agreement brings Google's Gemini AI models into Apple's ecosystem, fundamentally changing how the company approaches artificial intelligence on its devices.

What Changed in Apple's AI Strategy?

For years, Apple positioned itself as the privacy-first alternative to Google and other tech giants, promising that artificial intelligence would run directly on your device without sending data to distant servers. That vision remains central to Apple Intelligence, the company's personal AI system woven into iOS, iPadOS, macOS, watchOS, and visionOS. But the Gemini partnership reveals cracks in that strategy.

Apple Intelligence launched in October 2024 with iOS 18.1, operating on a two-tier architecture. Simple tasks like rewriting sentences or summarizing notes run entirely on-device using a roughly 3-billion-parameter model, which is a relatively small AI system optimized for speed. More complex requests route to Private Cloud Compute, Apple-designed servers running Apple silicon with cryptographic guarantees that data is processed but never stored or seen by Apple employees.

The Gemini deal disrupts this clean separation. The new Siri features in iOS 26.4 run on a 1.2-trillion-parameter architecture, a dramatically larger model that Apple is licensing from Google. While Apple states this runs on its own Private Cloud Compute servers with Gemini's model weights hosted by Apple rather than Google, the arrangement marks a departure from the company's original vision of keeping everything in-house.

Why Did Apple Need Google's Help?

The timing tells the real story. Apple's internal AI chief John Giannandrea departed almost exactly when the Gemini partnership was announced. The company had spent billions building an in-house AI team but couldn't ship a working Siri upgrade in two years. By early 2026, that failure became impossible to hide.

The financial terms underscore how serious the situation was. Bloomberg's Mark Gurman estimates the Gemini license costs Apple approximately $1 billion per year, though other reports citing Chinese tech outlets put the figure closer to $10 billion annually. Apple has not officially confirmed either number, but the scale of investment signals desperation rather than strategic choice.

Meanwhile, Apple Intelligence adoption has been strong among eligible users. According to Morgan Stanley data from April 2025, approximately 80% of eligible US iPhone users have used Apple Intelligence at some point. The infrastructure now handles roughly 1.5 billion Siri requests per day across 2.3 billion active Apple devices globally. The problem wasn't adoption; it was capability.

What Apple Intelligence Features Actually Work Right Now?

Not all of Apple Intelligence is struggling. Several features have matured into genuinely useful tools that shipped with iOS 26, launched at WWDC 2025. Understanding what works and what remains in development helps explain why the Gemini partnership became necessary.

• Writing Tools: Available across Mail, Notes, Messages, Safari, and third-party apps, these tools let you select any text and choose from Proofread, Rewrite, or Summarize options. Simple edits run on-device while complex rewrites route to Private Cloud Compute, and this feature works reliably enough that it quietly made Apple Intelligence worth enabling.
• Visual Intelligence: Point your camera at anything, and iOS identifies it, lets you search it, add calendar events, or ask ChatGPT about it. Screenshots now carry a triple-action bar for asking ChatGPT, image search, or adding to calendar, making this the most practically useful addition for everyday iPhone users.
• Live Translation: New in iOS 26, real-time two-way translation built into Messages, FaceTime, and Phone works without app switching or third-party services. For anyone regularly communicating across languages, this feature makes iOS 26 feel genuinely different.
• Image Generation Suite: Image Playground generates images from text or emoji prompts, now available as custom conversation backgrounds in Messages. Genmoji creates custom emoji from text descriptions, Clean Up removes unwanted objects from photos with AI-powered inpainting, and Memory Movies generates photo slideshows with music and transitions.
• Siri Enhancements: Type to Siri by double-tapping the bottom bar for silent interaction, and Siri now retains context across a session and can walk you through device settings step by step. The ChatGPT handoff is user-controlled and permission-gated, with Siri asking before sending anything to OpenAI.

What Features Are Still Missing?

The features that remain in development are the ones Apple promised in 2024 but couldn't deliver. Onscreen awareness and personal context, which would let Siri read your actual emails and calendar to answer complex questions, are still being worked on. These capabilities are central to Apple's original vision of a truly personal AI assistant that understands your life without uploading everything to the cloud.

The full chatbot-style Siri, internally called Apple Foundation Models v11, is expected to arrive with iOS 27 in fall 2026, likely previewed at the June 8 keynote. Bloomberg's Gurman reports it may run on Google's own cloud infrastructure for advanced queries, which would represent a significant departure from Apple's privacy architecture and a gap in its own messaging that hasn't been publicly addressed.

How to Understand What This Means for Your Privacy

• On-Device Processing: Simple tasks still run entirely on your iPhone or iPad without any data leaving your device, using the 3-billion-parameter model for fast, private processing of basic requests like proofreading or summarization.
• Private Cloud Compute: More complex requests route to Apple-designed servers running Apple silicon, with cryptographic guarantees that your data is processed but never stored or seen by Apple employees, and independent security researchers can audit these guarantees.
• Google Partnership Implications: The Gemini-powered Siri features run on Apple's Private Cloud Compute servers with Gemini's model weights hosted by Apple, not Google, though user data handling for the full iOS 27 Siri remains unclear and hasn't been publicly detailed.
• User Control: When Siri needs to hand off requests to ChatGPT or other services, it asks for permission first, giving you explicit control over when your data leaves Apple's infrastructure.

Google Cloud CEO Thomas Kurian confirmed the partnership at Google Cloud Next 2026, calling Google Apple's "preferred cloud provider." That phrase, used by Google executives rather than Apple, is the detail that should make privacy-conscious enterprise IT teams pause. It suggests a deeper relationship than Apple has publicly acknowledged.

Thomas Kurian

Apple also announced on May 19, 2026, a suite of AI-powered accessibility updates arriving later this year, including an enhanced VoiceOver that reads bills, photos, and personal documents in detail; Live Recognition on iPhone for real-time camera-based object identification; Voice Control powered by Apple Intelligence; on-device generated subtitles for uncaptioned video; and wheelchair eye-control integration for Vision Pro. These applications of AI have unambiguous value for users with disabilities.

"These features build on 40 years of accessibility innovation at Apple," stated Sarah Herrlinger, Senior Director of Global Accessibility Policy and Initiatives at Apple.

Sarah Herrlinger, Senior Director, Global Accessibility Policy and Initiatives, Apple Inc.

The Gemini partnership isn't a technology partnership in the traditional sense; it's an admission that Apple's internal AI capabilities couldn't meet the timeline and performance expectations the company had set. Whether the Google deal actually solves the problem is what WWDC 2026 will begin to answer. For now, Apple Intelligence remains a work in progress, and the company's pivot to Google's models signals that the original vision of keeping everything in-house was more aspirational than achievable.