Apple AI powered by Google Gemini models and Nvidia chips, but maintains distinct architecture

Apple AI Strategy Reveals Deep Google Gemini Integration

Apple executives confirmed at its annual Worldwide Developers Conference that the company's Apple AI capabilities rely significantly on Google Gemini models and Nvidia hardware, marking a strategic shift from building entirely proprietary systems 1

1

. The revelation came during a technical deep dive following the WWDC keynote, where Craig Federighi, Apple's SVP of Software Engineering, detailed how Apple's partnership with Google extends beyond what was initially disclosed in their January announcement 4

4

.

The company unveiled its third generation of Apple Foundation Models, a family of five foundation models custom-built in collaboration with Google 2

2

. These models power the redesigned Siri AI, which demonstrated the ability to check concert dates, set reminders, and provide directions in a fluid, conversational manner. Apple software SVP Craig Federighi emphasized the company's distinct approach, stating that "some appear to be racing forward, seemingly pursuing AI for the sake of AI, without clear regard for the people" 1

1

.

How Apple Foundation Models Differ From Google's Gemini Assistant

Despite using Google Gemini as the foundation, Apple executives were emphatic that Siri AI is not simply a rebadged version of Gemini Assistant. "The amount of the Google Assistant we use is none," Federighi clarified, explaining that Apple uses none of the Gemini models deployed to Google's customers, none of Google's client-side code, and no Google Search infrastructure 4

4

. Instead, Apple started with Gemini's foundation models, optimized and rebuilt them for Apple Silicon, and retrained them with proprietary data, weights, and guardrails 2

2

.

The architecture includes two on-device models: AFM 3 Core, a 3-billion-parameter dense model, and AFM 3 Core Advanced, a 20-billion-parameter model with sparse architecture that activates just 1 to 4 billion parameters depending on the request 3

3

. These on-device models run exclusively on Apple Silicon, with AFM 3 Core Advanced requiring an iPhone 17 Pro, iPhone Air, Macs with M3 and at least 12GB of RAM, or iPads with M4 3

3

.

Private Cloud Compute Extends to Google Servers With Nvidia GPUs

The most significant architectural change involves AFM Cloud Pro, Apple's most capable server-based model designed for agentic tool use and complex reasoning tasks 4

4

. To run this model, Apple worked with both Google and Nvidia to extend Private Cloud Compute infrastructure to Nvidia GPUs hosted in Google's cloud 5

5

. This marks the first time Apple has officially confirmed that Apple Intelligence features will run on Nvidia chips outside Apple's own data centers 1

1

.

Apple's core PCC requirements remain unchanged: stateless computation, enforceable guarantees, no privileged runtime access, non-targetability, and verifiable transparency 5

5

. The implementation leverages Nvidia Confidential Computing technology called "ambiguous confidential compute," which prevents Nvidia GPUs from reading the contents of Apple's servers 4

4

. Apple maintains a cryptographically verifiable ledger of all Google Cloud hardware that is part of the PCC fleet to mitigate supply chain attack risks 5

5

.

AI Model Privacy and Security Architecture Explained

The System Orchestrator serves as the cornerstone of Apple's privacy architecture, routing queries to the appropriate model based on complexity and required personal context 4

4

. Two of the four Apple Foundation Models run entirely on-device, ensuring absolute data protection since information never leaves the device [2](https://9to5mac.com/2026/06/11/siri-ai-is-powered-by-gemini-models-but is-not-gemini-what-does-that-mean/). The next two cloud-based models run on Apple Silicon chips within Apple's own Private Cloud Compute servers, where no data is retained or exposed to either Apple or Google—a claim that is independently verifiable by security researchers 2

2

.

For queries involving current events, responses come from Apple's own World Knowledge Service, which the company has been building for several years, rather than relying on Google Search 4

4

. PCC on Google Cloud binaries will be available for public inspection, with Apple planning to provide research tooling and access to live PCC nodes through its Apple Security Bounty Program 5

5

. However, PCC on Google Cloud is not fully implemented, and Apple plans to gradually add the complete set of protections throughout beta testing 5

5

.

What Apple's AI Advancements Mean for Users

Apple's approach represents a calculated middle path between building entirely proprietary AI systems and fully outsourcing capabilities to established AI leaders. The company chose not to spend billions on infrastructure and the biggest, most advanced models, instead focusing its message on privacy advantages and convenience 1

1

. All four models optimized for Apple Silicon were "trained using proprietary data with reinforcement learning and refined using outputs from Gemini frontier models," according to AI VP Amar Subramanya 4

4

.

The partnership allows Apple to deliver competitive AI capabilities while maintaining control over user experience and data handling. Security researchers can verify the verifiable transparency claims independently, meaning users don't have to rely solely on Apple's assurances 2

2

. As Apple continues refining PCC on Google Cloud during beta testing, the technology industry will be watching closely to see whether this hybrid approach successfully balances performance, privacy, and practical deployment at scale.🟡선을)