Substrate
technology

Apple Extends Siri AI with Google Gemini Models and Nvidia Hardware While Expanding Private Cloud Compute Privacy Safeguards

Apple detailed at WWDC how its new Siri AI and Apple Intelligence features will run partly on Google Cloud servers while maintaining privacy protections through an expanded Private Cloud Compute system.

Ars Technica
1 source·Jun 9, 9:05 AM·2m read
Apple Extends Siri AI with Google Gemini Models and Nvidia Hardware While Expanding Private Cloud Compute Privacy SafeguardsThe Verge
Audio version
Tap play to generate a narrated version.
Developing·Limited corroboration so far. This page will refresh as more sources emerge.

Apple announced that its Siri AI upgrade will use Google’s Gemini language models and run on Nvidia hardware in Google servers, while keeping all privacy guarantees that previously applied only to Apple-controlled hardware. The company made the confirmation June 8 at its Worldwide Developers Conference in Cupertino, California.

” Most Apple Intelligence devices will receive AFM 3 Core, a Gemini-based model co-developed by Apple and Google, in the 2026 operating system releases.

Devices with at least 12 GB of RAM and recent chips—M3 and newer Macs, M4 and newer iPads, A19 Pro iPhones—will instead run AFM 3 Core Advanced, which uses extra hardware and device storage to improve dictation and power Siri’s more expressive voice.

For complex queries, devices will contact three cloud models also co-developed by the two companies: AFM 3 Cloud for general use, ADM 3 Cloud for image generation, and AFM 3 Cloud Pro for agentic tool use and complex reasoning. AFM 3 Cloud and ADM 3 Cloud run on Apple silicon in Apple data centers; AFM 3 Cloud Pro runs on Google-owned Nvidia hardware.

To extend its privacy architecture to third-party servers, Apple introduced a new version of Private Cloud Compute. The system uses Nvidia’s Confidential Computing, Intel’s Trust Domain Extensions, and Google’s Titan security chip. Apple maintains a cryptographically verifiable, append-only ledger of all Google Cloud hardware in the PCC fleet, and devices will only trust Apple-signed software on those servers.

Google Cloud servers do not yet match all protections of Apple’s own Private Cloud Compute servers, but Apple said it will gradually add the remaining safeguards during the summer preview period. An on-device System Orchestrator decides which model to use and ensures only the minimum data required for a query leaves the device.

Federighi said the architecture is designed so that “PCC itself, by design from the ground up, is going to vaporize any record of that data the moment after it answers your question… This is not stored.

Siri AI and the other new Apple Intelligence features will ship this fall in iOS 27, iPadOS 27, macOS 27 Golden Gate, and related operating systems. Developer betas are available now; a more stable public beta is scheduled for July. Apple previously relied on its own server hardware for Private Cloud Compute.

The company said it had reached the limits of on-device hardware for large language and reasoning models and needed additional capacity without building its own large data-center fleet.

Transparency

1 source · single source
CorroborationModerate · 1 source

Story details