Apple will run its redesigned Siri on Google's servers starting late 2026. The company shifts its voice assistant's cloud infrastructure to a third party for the first time. Bloomberg reported the arrangement alongside details of Campos, the code name for Apple's project. Campos transforms Siri from command queries into open conversations powered by Gemini AI technology.

Apple plans a two-stage rollout. iOS 26.4 delivers Gemini-powered AI features to Siri in mid-2026. iOS 27 follows later that year with Campos. Users will switch between typing and speaking mid-conversation. Apple will unveil the system at WWDC 2026 as the flagship AI upgrade for iOS and macOS 27.

Users could start a question by typing on their phone, then finish speaking to their watch. Developers will build new voice interfaces around the conversational model. Teams must decide whether to rebuild existing Siri Shortcuts or adapt current implementations.

The infrastructure shift raises questions about Apple's privacy positioning. Apple announced Private Cloud Compute in June 2024. The proprietary architecture uses Apple silicon servers with stateless processing, which erases data immediately after handling requests. The design prevents Apple from accessing user data during cloud AI operations.

Running Campos on Google's infrastructure contradicts that approach. Users will send voice data to Google's data centers, requiring privacy advocates to evaluate Google's handling of Apple user requests. Apple has stored encrypted iCloud data chunks on Amazon S3 and Google Cloud Platform since at least 2016, according to its iOS Security Guide published in February 2018.

Apple builds its model on top of Gemini rather than licensing Google's product wholesale. Bloomberg reported a roughly $1 billion annual licensing figure for the multi-year agreement. Reuters and The Verge confirmed Apple constructs foundation models—the underlying AI architecture—using Gemini technology as the base layer. This preserves some architectural control while outsourcing server operations.

Apple constructs its conversational layer while Google operates the infrastructure. The arrangement lets Apple customize behavior while Google scales compute resources.

Apple has not revealed whether Campos will expose APIs for third-party integration. Teams building voice features lack information about migration paths. Existing Siri Shortcuts developers wait for guidance on whether current implementations will function or require complete rebuilds. Companies must plan for two scenarios: either Apple provides compatibility layers or developers rewrite voice experiences from scratch.

WWDC 2026 will reveal whether Apple can maintain privacy promises while running on Google infrastructure. The event should clarify interaction patterns, offline capability, and API availability for developers. Apple must explain how it reconciles third-party server dependence with its privacy-first messaging. Until then, the infrastructure trade-offs behind Apple's platform direction remain undefined.