Apple says its AI is still private, even when it's running on Google's servers

2026-06-09 · Source: AI - Ars Technica · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Cybersecurity & Data Privacy · Depth: Intermediate, short

Summary

Apple's new "Siri AI" and "Apple Intelligence" features will integrate Google's Gemini language models and run on Nvidia hardware within Google's servers, a significant shift from Apple's traditional on-device or Apple-controlled server processing. Despite this, Apple asserts it maintains its stringent user privacy promises. The company addresses hardware limitations for large language models by extending its Private Cloud Compute (PCC) system to third-party infrastructure. This is achieved through a multi-layered security approach, incorporating Nvidia's Confidential Computing, Intel's Trust Domain Extensions, and Google's Titan security chip. Additionally, Apple maintains a cryptographically verifiable ledger of Google Cloud hardware and ensures only Apple-signed software runs on these servers. An on-device "System Orchestrator" minimizes data sent off-device, and PCC is designed to ensure data transience, vaporizing records immediately after query resolution. These features will launch with iOS 27, iPadOS 27, and macOS 27 Golden Gate this fall.

Key takeaway

For AI Architects evaluating hybrid cloud strategies for sensitive data, Apple's approach demonstrates a viable path to extend privacy guarantees beyond proprietary infrastructure. You should investigate integrating confidential computing technologies like Nvidia's Confidential Computing or Intel's Trust Domain Extensions. Prioritize on-device data orchestration to minimize off-device transfers and design cloud processes for strict data transience, ensuring no persistent records. This model allows scaling AI capabilities without compromising user trust.

Key insights

Apple extends its privacy architecture to third-party clouds using hardware-backed confidential computing and strict data transience.

Principles

• On-device orchestration minimizes data exposure.
• Data processed in cloud must be transient.
• Hardware-backed confidential computing is crucial.

Method

Apple's Private Cloud Compute (PCC) on Google Cloud uses Nvidia's Confidential Computing, Intel's Trust Domain Extensions, and Google's Titan chip. An on-device System Orchestrator selects models and data, sending minimal, transient data to PCC for processing.

In practice

• Implement on-device data filtering.
• Design for transient cloud data processing.
• Utilize confidential computing technologies.

Topics

• Apple Intelligence
• Private Cloud Compute
• Confidential Computing
• Data Privacy
• Hybrid Cloud AI
• Gemini Language Models

Best for: CTO, VP of Engineering/Data, Executive, AI Architect, Director of AI/ML, AI Product Manager

Related on AIssential

• Apple’s new Siri AI overhaul is here (sort of) — Apple's Siri AI integrates on-device and cloud processing for enhanced context, privacy, and system-wide actions.
• Perplexity announces hybrid AI system that decides what runs locally or in the cloud — Perplexity's hybrid AI orchestrator intelligently routes tasks between local and cloud models for optimal privacy and efficiency.
• Google and Blackstone partner on TPU-powered data centre capacity — Google and Blackstone's CaaS venture offers TPU-powered AI compute as core infrastructure, diversifying enterprise access beyond public clouds.
• Apple bets cheaper AI will woo small developers — Apple is reducing AI infrastructure costs for small developers to foster innovation and attract new talent.
• Building a Zero-Trust Architecture for Confidential AI Factories — Confidential computing with TEEs and CoCo enables secure, zero-trust AI factories by encrypting data and models during execution.
• IBM’s enterprise AI strategy makes trust and control the production test — Governed enterprise AI, integrated with hybrid cloud and robust security, is essential for production-scale trust and value.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI - Ars Technica.