Apple partnering with Google and Nvidia for most advanced AI model

Apple on Monday revealed what it’s been working on in artificial intelligence at its annual Worldwide Developers Conference in Cupertino, Calif.

WWDC showed off demos of its redesigned Siri, which can speak back and forth with the user, a major improvement over previous versions of the assistant. In a demo, Siri was able to check concert dates, set a reminder to buy tickets, and even get directions to pick up a friend on the way to the concert venue.

But the announcement also highlighted that Apple has taken a different strategy to many of its Silicon Valley rivals, choosing not to spend billions on infrastructure and the biggest, most advanced models, and instead focusing its message to potential customers on privacy advantages and convenience.

Apple executives highlighted the difference in remarks on Monday.

“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,” said Apple software SVP Craig Federighi in the launch announcement.

But it turns out two of the traditional AI leaders, Google and Nvidia, are helping Apple out with its most advanced model, called Apple Foundation Model Cloud Pro, Apple executives told media in a talk at its headquarters on Monday.

While Apple and Google announced their partnership for Apple Intelligence in January, this is the first time that the company has officially confirmed that some of its Apple Intelligence features will run on Nvidia chips.

Apple AI executive Amar Subramanya said that AFM Cloud Pro is comparable to Google’s Gemini frontier models. It will run in the cloud on Nvidia GPUs, which are part of Apple’s Private Cloud Compute infrastructure, Apple officials said.

“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,” Subramanya said.

VP of software Sebastian Marineau-Mes said that Apple wanted to use Nvidia’s latest chips, but Apple wanted the chips to be configured in a more private way, where they couldn’t read what was on the servers.

Marineau-Mes said that recent Nvidia improvements, such as a technology called “ambiguous confidential compute,” allowed Apple and Google to build a system that met its standards.

“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.

Apple is distinguishing itself from companies that have more heavily invested in AI by emphasizing that its software is more private — the company isn’t collecting as much data as web-based AI such as OpenAI’s ChatGPT or Anthropic‘s Claude — and that it is using its access to locally-stored user information like a calendar or text messages to personalize AI features.

The tech talk on Monday was held so that Federighi and his lieutenants could discuss how Apple built Siri AI and its Apple Intelligence layer and how it differs from AI that consumers might already be familiar with.

Apple executives outlined an architecture for the software in which Apple’s operating systems and software have a piece of software called a system orchestrator that routes any AI query to an appropriate model — either on the device or in the cloud — depending on how much computing power and personal data it needs.

The system orchestrator is “key to the privacy architecture of our entire system,” said Federighi.

The talk also provided some more details on the Apple-Google partnership.

Federighi said that Apple Intelligence, the AI software built into the company’s devices, uses Apple’s own models, not the same Google Gemini available to the public, as many people in the tech industry expected when the two companies announced a partnership in January. He also said it wasn’t using Google’s off-the-shelf cloud infrastructure.

Federighi said that Google’s technology was used to help build Apple’s own models — specifically, “third-generation” AFM models for the cloud announced on Monday that are designed and turned to run on Apple’s chips.

“These four models that we just talked about — AFM Core, Core Advanced Cloud, and Cloud Image —all of these are custom built for Apple Silicon, trained using proprietary data with reinforcement learning and refined using outputs from Gemini frontier models,” Subramanya said.