Apple partners with Google to power Siri AI

Apple and Google announced a multi-year collaboration in which the next generation of Apple Foundation Models will be based on Google's Gemini models and cloud technology, per a joint statement published on Google's blog (Google Blog, Jan 12, 2026). At WWDC 2026, Apple demonstrated a renamed assistant, "Siri AI," showing more conversational and task-oriented demos across iOS, iPadOS, macOS, watchOS, visionOS and tvOS (NBC News, Wired). CNBC reports Apple described a cloud model named Apple Foundation Model Cloud Pro and said it will run in cloud environments on Nvidia GPUs, with Apple executives comparing its capabilities to Google's frontier models (CNBC). Apple has emphasised that Apple Intelligence features will continue to run on devices and on its Private Cloud Compute system while maintaining its stated privacy standards (Google Blog; Wired).

Editorial analysis: Reporting indicates the announced stack is hybrid, combining on-device processing, Apple's Private Cloud Compute (PCC), and Google-hosted model inference. Sources show Apple intends to use Gemini as a foundation for Apple's AFM derivatives (Google Blog), and CNBC quotes Apple saying AFM Cloud Pro will rely on Nvidia GPU infrastructure in the cloud. Independent commentary from a cryptography researcher notes Apple described PCC as encrypting data from device to server and purging it after a response, but the researcher describes Apple's expansion of PCC to include Google hardware as technically vague and potentially reliant on Google's existing infrastructure (Cryptography Engineering blog, June 9, 2026).

The collaboration combines Apple's control of the device layer and privacy branding with Google's frontier model capabilities. Public reporting frames the move as Apple choosing an established model provider over alternatives that were reportedly considered, including OpenAI and Anthropic (The Verge). For practitioners, the partnership matters because it changes where heavy inference is likely to run, which affects latency, cost, and the operational requirements for secure transit and key management between device, Apple datacenters, and third-party cloud environments.

Editorial analysis: Apple reiterates a privacy posture by naming on-device execution and PCC as control points (Google Blog; Wired). Independent reporters and experts point out the tension between using centralized, high-capability models and Apple's promise that "your data never leaves Apple's hardware" under PCC, especially where Google operates parts of the cloud stack (Cryptography Engineering blog). This raises concrete engineering questions about encryption-in-transit, key custody, attestation of remote hardware, and auditability when third-party cloud providers host model execution.

Observers should track several indicators that are straightforward to verify: official technical documentation or whitepapers describing PCC's cryptographic protocols and key custody; published details on where Gemini instances will execute and the boundaries of Apple-controlled infrastructure; latency and cost benchmarks for AFM Cloud Pro on Nvidia GPUs; and any third-party audits or attestation mechanisms that Apple publishes for cross-cloud inference. Also watch developer-facing SDKs and API contracts Apple releases for Siri AI integration, which will reveal the practical data flows and developer security responsibilities.

Editorial analysis: The Apple-Google arrangement is a pragmatic industry move that layers frontier models behind a privacy-centric product story. Engineers building integrations should treat reported claims about PCC and encryption as hypotheses to validate against forthcoming technical documentation and audits. The deployment model, hybrid device plus cloud using third-party GPUs, implies teams will need to design for mixed trust boundaries, instrumented telemetry for interaction tracing, and careful data-minimization to align with both user privacy expectations and operational constraints.