Apple confirms Siri AI will run on Google servers, citing privacy safeguards

Apple has confirmed that its upcoming Siri AI feature, which leverages Google’s Gemini language models, will operate on Nvidia hardware within Google’s cloud infrastructure. The announcement, made during the company’s Worldwide Developers Conference, outlines a shift in how the tech giant handles complex queries that exceed the capabilities of on-device processing. To address concerns regarding user privacy, Apple introduced a new iteration of its Private Cloud Compute system designed to function on third-party hardware while maintaining strict security protocols.

Craig Federighi, Apple’s senior vice president of Software Engineering, described the architecture as a traditional chatbot setup where the client application on the user’s device communicates with cloud-based models. He emphasised that the company utilises “none” of the Google system in a way that grants access to the underlying infrastructure. The system relies on an on-device “System Orchestrator” to determine which model is required for a query, ensuring that only minimal necessary data is transmitted off-device.

The model distribution is segmented based on hardware capability. Simpler tasks are handled by AFM 3 Core or AFM 3 Core Advanced models running locally on Apple silicon. For more sophisticated requests, the system accesses cloud-based models including AFM 3 Cloud and ADM 3 Cloud, which run on Apple’s own servers. The most complex tasks, involving agentic tool use and advanced reasoning, are processed by the AFM 3 Cloud Pro model, which operates on Google-owned Nvidia hardware.

To secure data processed on these external servers, Apple is employing hardware-level security measures including Nvidia’s Confidential Computing, Intel’s Trust Domain Extensions, and Google’s Titan security chip. The company maintains a cryptographically verifiable, append-only ledger of all Google Cloud hardware within the Private Cloud Compute fleet. Apple devices are configured to trust only software signed by the company, and Federighi stated that the architecture is designed to “vaporize” any record of data immediately after it answers a question, ensuring the information remains transient and is not stored.

Siri AI and other Apple Intelligence features are scheduled to launch this autumn as part of iOS 27, iPadOS 27, and macOS 27 Golden Gate. Developer betas are currently available, with public beta versions expected to be released in July. This architectural adjustment allows Apple to access greater compute capacity for its AI services without committing to the massive internal data centre buildouts that previous scaling attempts would have required.