Apple deploys 1.2T-parameter Gemini model on Nvidia Blackwell GPUs for rebuilt Siri

Apple announced at WWDC 2026 that the rebuilt Siri runs on a custom 1.2-trillion-parameter model based on Google's Gemini technology, hosted on Google Cloud servers powered by Nvidia Blackwell B200 GPUs. The deal reportedly costs Apple roughly $1 billion per year.

Three-tier privacy architecture

Apple routes Siri queries through three layers:

1. On-device: Simple tasks stay local using Apple's own models
2. Private Cloud Compute: Moderately complex requests go to Apple's servers
3. Google Cloud: The heaviest reasoning tasks route to Google infrastructure

At each tier, Apple claims queries are anonymised and tokenised so neither Apple staff nor Google can link requests to individual users. Software chief Craig Federighi stated: "Your requests are completely private to you. They're never stored. They're never accessible to anyone."

The contract with Google reportedly bars the company from training future models on Apple user data. Nvidia's confidential computing feature encrypts data while being processed on the Blackwell GPUs, adding hardware-level protection.

No independent audit of the Google Cloud tier has been published, and contractual bans on training can be renegotiated in future deals.

Five new Apple Foundation Models

Apple unveiled the third generation of its Apple Foundation Models (AFM), a family of five models distilled from Gemini:

• AFM Core
• AFM Core Advanced
• AFM Cloud
• AFM Cloud Pro
• AFM Cloud Image

The most powerful, AFM Cloud Pro, offers quality "similar" to Google's frontier Gemini models, according to AI VP Amar Subramanya, though no independent benchmark has confirmed the comparison. All five are custom-built for Apple Silicon and trained with proprietary data and reinforcement learning.

Context and timing

At WWDC 2025, Federighi and marketing chief Greg Joswiak dismissed the idea of a "bolted-on chatbot." When asked what changed, Federighi said: "We see Siri not as a separate chatbot, but rather as an integral but conversational tool that you use in the moment."

Apple settled a $250 million class action lawsuit last month over marketing AI features in 2024 that were not ready when the iPhone 16 launched. Siri engineering lead Mike Rockwell acknowledged that previous attempts to revamp the assistant "didn't meet Apple's standards."

What this means

The deal creates a dependency on Google—simultaneously Apple's biggest rival in mobile operating systems and its largest source of search revenue—for frontier-class AI capabilities. Apple avoided building large-scale AI infrastructure from scratch but now relies on a competitor's models and cloud infrastructure for core iOS functionality.

The privacy architecture depends on contractual agreements, hardware encryption, and Apple's assurances. Without independent audits of the Google Cloud tier, users must trust that the three-tier system performs as described. The features ship in September, when real-world implementation and user trust will be tested.