Particle.news

Apple Rebuilds Siri Using Licensed Google Gemini for Cloud Reasoning

The company will send compute‑heavy queries to a custom Gemini deployment on Google Cloud with confidential compute to deliver faster, more capable conversational AI and preserve privacy controls.

Overview

  • Apple plans to preview the rebuilt Siri at WWDC on June 8, open a public beta this summer, and ship the finished assistant with iOS 27 and new iPhone hardware in September.
  • The new design splits work: smaller models will run on users’ devices for routine tasks while complex, reasoning queries will be routed to a licensed version of Google’s Gemini running on Google Cloud.
  • Reports say Apple agreed to pay roughly $1 billion a year for the Gemini model after finding its Private Cloud Compute and in‑house servers could not meet real‑time inference needs.
  • The rollout is expected to bring a standalone Siri app plus features such as conversational history, file and image uploads, screen awareness, Dynamic Island integration, and multi‑app chaining.
  • Apple is framing confidential compute, on‑device models, and auto‑delete controls as privacy safeguards but the move creates a new dependency on Google and Nvidia that could reshape Apple’s long‑held ‘own the stack’ approach.