Apple confirmed at WWDC 2026 on June 8 that it is licensing a custom 1.2-trillion-parameter Gemini model from Google at approximately $1 billion per year to rebuild Siri from scratch. The deal, announced on stage by Tim Cook in what was his final WWDC keynote as CEO, is the largest single payment Apple has made to a competitor in AI and the clearest public acknowledgment that Apple could not build a competitive frontier assistant on its own.
The financial terms put the cost of catching up in concrete terms. Bloomberg framed the keynote as “a confession dressed up as a product launch.” Apple spent years and billions of dollars on custom silicon, on-device inference, and the Private Cloud Compute infrastructure it unveiled in 2024. None of it produced a large language model capable of competing with Gemini or GPT-4o at the frontier. The $1 billion annual license is what that gap costs.
The rebuilt Siri ships with a dedicated app featuring a chatbot-style interface, a system-wide “Search or Ask” gesture, and Dynamic Island integration on iPhone 16 and newer. With explicit user permission, the new Siri can access emails, photos, messages, calendar data, and files for cross-app multi-step tasks. Lighter Apple Intelligence features stay on-device; heavier reasoning routes to the licensed Gemini backend.
The second announcement is structurally more significant for the AI industry. Apple introduced Multi-AI Extensions, a system that lets users designate a third-party AI as their default conversational assistant on iPhone. Claude is one of the launch options, making it an iPhone default for the first time. The Extensions interface is built to let multiple AI providers compete for the user role inside Apple’s surface, which has been closed to external AI since Siri launched in 2011.
The timing lands alongside two other platform-level shifts. Microsoft last week announced Scout, a multi-provider AI routing layer that lets enterprise customers direct queries across Copilot, Claude, and Gemini without locking into a single model. Apple also opened iMessage to Poke, the AI-native messaging startup, the week before WWDC. Three distribution moves in two weeks all point to the same structural conclusion: the major platform companies are accepting that no single AI model wins, and are repositioning themselves as routers rather than exclusive AI endpoints.
For AI labs, Multi-AI Extensions is a distribution opportunity with no historical precedent on iOS. Anurag Rana, senior equity analyst for software and IT services at Bloomberg Intelligence, noted that roughly 1 billion iPhones currently in use do not support Apple Intelligence because it requires iPhone 15 Pro or later. The Extensions system runs on a broader hardware base, which means the addressable audience for Claude, Gemini, and any future Extensions partner is larger than Apple’s own AI footprint today.
The management context sharpens the stakes. Cook hands the CEO title to hardware chief John Ternus on September 1 and moves to executive chairman. Ternus will inherit a Siri architecture built on a competitor’s model, a platform newly open to rival AI providers, and a Wall Street audience that has spent two years demanding clarity on Apple’s AI direction. Whether the Gemini license includes exclusivity terms has not been disclosed. That distinction matters: a non-exclusive arrangement would leave the door open for Apple to add OpenAI or Mistral as future Siri backends.
Apple announced iOS 27 beta the afternoon of the keynote, completing an OS-family rollout across iOS, iPadOS, macOS, watchOS, and visionOS.
Teams building on Claude or evaluating Gemini for enterprise deployment should treat the Multi-AI Extensions launch as a distribution signal worth watching in Q3; if Apple discloses Extensions install rates at its September event, those numbers will set the benchmark for what iOS AI distribution actually looks like at scale.
CNN (cnn.com), 2026-06-08. Additional sourcing from the Apple WWDC 2026 keynote, June 8.
