Particle.news
Download on the App Store

Apple Rebuilds Siri With a Hybrid On‑Device and Cloud AI Stack

Combining on‑device models with Google Cloud using Nvidia GPUs under privacy safeguards, Apple plans a staged, hardware‑gated rollout for Siri AI.

Overview

  • Apple unveiled the rebuilt Siri, called Siri AI, at WWDC and released developer betas for iOS 27 with a public beta due next month and a broad consumer release expected this autumn.
  • The system uses a hybrid architecture: many requests run on device for speed and privacy while heavier inference is routed to Apple’s Private Cloud Compute and to Google Cloud running on Nvidia Blackwell GPUs.
  • Apple says it will rely on confidential‑compute techniques to protect data sent to cloud GPUs, but reporting has raised questions about how that encryption works and about a reported near‑$1 billion commercial arrangement with Google.
  • Key Siri AI features — such as cross‑app actions, visual intelligence, image generation and extended message or calendar tasks — will be limited to newer hardware and will launch first in English in the U.S., with the EU and China excluded initially.
  • Analysts gave mixed responses: some praised Apple’s privacy‑forward, developer‑first play while investors worried about timing, limited regional availability and ongoing costs for cloud model use; the rollout will test whether Apple can make AI a system‑level advantage rather than a standalone product.