Apple's AFM Cloud Pro Model Runs on Nvidia GPUs in Google Cloud, Execs Confirm

Apple executives confirmed at its Worldwide Developers Conference on Monday that the company's most advanced AI model, Apple Foundation Model Cloud Pro (AFM Cloud Pro), runs on Nvidia GPUs deployed within Google's cloud infrastructure. This marks the first official confirmation of Nvidia's role in Apple Intelligence, though the Apple-Google partnership was announced in January 2026.

Technical Architecture Details

According to Apple AI executive Amar Subramanya, AFM Cloud Pro is "comparable to Google's Gemini frontier models" in capability. The model runs as part of Apple's Private Cloud Compute infrastructure, leveraging Nvidia's GPUs configured with enhanced privacy protections.

Apple VP of software Sebastian Marineau-Mes explained the company wanted access to Nvidia's latest chip technology but required custom configurations to prevent the servers from accessing user data. "We wanted to avail ourselves of the latest technology from Nvidia, and so we set out to extend private cloud compute to third-party cloud," Marineau-Mes said. Recent Nvidia improvements, specifically "Nvidia confidential compute" technology, enabled this privacy-preserving arrangement.

Apple's Multi-Model System

Apple disclosed it operates four distinct AFM models: AFM Core, Core Advanced, Cloud, and Cloud Image. All four models are custom-built for Apple Silicon chips and trained using proprietary data. However, the models were "refined using outputs from Gemini frontier models," according to Subramanya.

The system uses what Apple calls a "system orchestrator" — software that routes AI queries to the appropriate model (on-device or cloud-based) depending on computational requirements and data access needs. Software SVP Craig Federighi described this orchestrator as "key to the privacy architecture of our entire system."

Strategic Positioning

Apple is explicitly differentiating itself from AI competitors by emphasizing privacy over raw model capability. "Some appear to be racing forward, seemingly pursuing AI for the sake of AI, without clear regard for the people — all of us — that it's ultimately meant to serve," Federighi said during the keynote.

The company is not using Google's off-the-shelf cloud infrastructure or public Gemini models, despite the partnership. Instead, Google's technology helped Apple build and refine custom models optimized for its own hardware ecosystem.

Pricing for AFM Cloud Pro access was not disclosed. Apple demonstrated improved Siri capabilities at WWDC, including conversational interactions, context awareness across calendar and messaging apps, and multi-step task completion.

What This Means

Apple's approach represents a hybrid strategy: building proprietary models while leveraging Google's frontier AI and Nvidia's compute infrastructure. The disclosure confirms Apple is not attempting to match competitors' infrastructure spending, instead focusing on privacy-preserving cloud compute and tight integration with its device ecosystem. The use of Gemini outputs to refine Apple's models suggests the company is using distillation or similar techniques rather than building frontier models from scratch. This could allow faster deployment while maintaining differentiation through privacy architecture and on-device capabilities.