Apple’s AI Bet: Why It’s Leaning on Google and Nvidia Instead of Building Its Own Data Centers

At its Worldwide Developers Conference in Cupertino on Monday, Apple peeled back the curtain on its AI strategy—and it looks nothing like what its Silicon Valley neighbors are doing. While rivals pour billions into massive data centers and sprawling models, Apple is taking a leaner, privacy-first approach. The centerpiece is a redesigned Siri that can hold a conversation, check concert dates, set reminders, and even route you through traffic to pick up a friend—all without hoarding your data.

But here’s the twist: Apple’s most advanced model, the Apple Foundation Model Cloud Pro (AFM Cloud Pro), won’t run on Apple-owned hardware alone. For the first time, the company confirmed it’s using Nvidia GPUs hosted in Google’s cloud to power this model. Apple AI executive Amar Subramanya told reporters that AFM Cloud Pro is on par with Google’s Gemini frontier models. “We work with both Google and Nvidia to extend our private cloud compute infrastructure to Nvidia GPUs in Google’s cloud, while maintaining Apple’s unmatched privacy guarantees,” he said.

The key enabler? A new Nvidia technology called “ambiguous confidential compute,” which keeps server data unreadable—even by Apple. VP of software Sebastian Marineau-Mes noted that Apple wanted Nvidia’s latest chips but needed them configured in a way that preserved user privacy. That technical hurdle, he said, has now been cleared.

Apple’s software SVP Craig Federighi emphasized that the company isn’t chasing AI for AI’s sake. “Some appear to be racing forward without clear regard for the people it’s meant to serve,” he said. Instead, Apple’s system orchestrator—a piece of software baked into its operating systems—routes each AI query to the right model, either on-device or in the cloud, depending on the computing power and personal data required. Federighi called it “key to the privacy architecture of our entire system.”

Notably, Apple Intelligence doesn’t use Google’s off-the-shelf Gemini model or standard cloud infrastructure. Instead, Google’s technology helped Apple build its own third-generation AFM models, trained on proprietary data and refined using outputs from Gemini frontier models. “These models are custom built for Apple Silicon,” Subramanya said.

The message is clear: Apple is betting that privacy, not raw scale, will win the AI race.