At the WWDC 2026 opening event, Apple presented an updated artificial intelligence strategy that prioritises user privacy and improved user experiences over raw compute power. The company also showcased a redesigned Siri capable of more natural, continuous conversation: in the demo Siri searched for concert dates, set a reminder to buy tickets and planned a route to pick up a friend on the way to the venue.
AFM Cloud Pro developed with Google and Nvidia
One of the notable announcements at WWDC was that two major industry players, Google and Nvidia, contributed to the development of Apple’s most advanced AI model, the Apple Foundation Model Cloud Pro (AFM Cloud Pro). Apple executives said in a post-event background briefing that both companies’ technologies were used during AFM Cloud Pro’s development.
Amar Subramanya, Apple’s head of AI development, said AFM Cloud Pro’s capabilities can be compared with Google Gemini’s more advanced models. The system runs in a cloud environment on Nvidia GPUs that form part of an expanded Private Cloud Compute infrastructure hosted in Google’s cloud. Subramanya emphasised that Apple extended its Private Cloud Compute with Nvidia accelerators in Google’s cloud while preserving Apple’s privacy guarantees.
Privacy-first approach and “ambiguous confidential compute”
Sebastian Marineau-Mes, Apple’s software engineering vice president, said Apple initially intended to use Nvidia’s most advanced AI accelerators but only in a way that meets Apple’s stringent privacy requirements. He noted that Nvidia’s recent developments — including a technology referred to as “ambiguous confidential compute” — made it possible to build infrastructure in which the operators of the servers cannot access the processed data. As a result, Apple extended its Private Cloud Compute system to external cloud infrastructures.
Apple devices do not run Gemini
In the background discussion Craig Federighi, Apple’s senior vice president of software engineering, stressed that Apple Intelligence does not run the publicly available Gemini models on devices. He also made clear that the system is not built on Google’s general-purpose cloud infrastructure; Google’s technology primarily contributed to the development of Apple’s own models.
Subramanya added that the third-generation AFM models presented — including AFM Core, Core Advanced Cloud and Cloud Image models — were specifically designed and optimised for Apple Silicon processors. These models were trained on proprietary datasets, improved via reinforcement learning, and fine-tuned using outputs from high-end Gemini models.
System orchestrator and local processing
In a technical demonstration following the keynote, Craig Federighi and colleagues described how Apple Intelligence works. A central element is the so-called system orchestrator, a control layer that routes AI requests to the appropriate model and decides whether a task can be executed locally on the device or requires cloud-based processing with greater compute resources. Federighi called the system orchestrator a key part of the platform’s privacy architecture.
How Apple’s approach differs and EU availability
While companies such as OpenAI and Anthropic increasingly rely on large volumes of online data to train their models, Apple emphasises that it collects far less user data and leverages personal information stored on users’ devices — such as calendar entries, messages and other local data — to personalise AI features while preserving privacy. Federighi remarked that some players seem to be advancing AI “for its own sake,” with less attention to the people the technology is meant to serve.
Apple also said that some of the features demonstrated at the keynote will not be fully available to users in the European Union immediately; EU users should expect an optimised system, improved search and image-sharing, and better parental controls in the interim, with the full release likely arriving around September alongside new product announcements.
Conclusion
The WWDC 2026 announcements show Apple’s intent to build a privacy-focused, device-first AI platform while increasingly relying on technologies from leading AI-industry companies behind the scenes. The company aims to combine on-device processing and data protection with cloud-based capabilities — running AFM Cloud Pro on Nvidia GPUs within an expanded Private Cloud Compute environment — to deliver advanced AI features without compromising its stated privacy principles.
