![]() |
From left to right: Amar Subramanya, Mike Rockwell, Sebastien Marineau-Mes, and Craig Federighi. Photo: Bloomberg . |
At WWDC 2026, Apple showcased the latest features of its improved Siri. Called Siri AI, this assistant can interact with users, extract information from any app, and operate seamlessly across devices.
The event also showed that Apple is taking a different approach. Instead of racing to invest billions of dollars in advanced hardware infrastructure, Apple focuses on usability and privacy when introducing products.
"Some parties seem to be acting recklessly. They're pursuing AI solely for technological gain without a clear concern for the users, the very people this technology is meant to serve," said Craig Federighi, Vice President of Software Engineering, at the event.
Apple has admitted this for the first time.
In fact, Google and Nvidia, leading companies in the AI field, are supporting Apple in developing its most advanced model. Called the Apple Foundation Model (AFM) Cloud Pro, this information was shared by company leaders at a press event on June 8th.
While Apple and Google announced their partnership in January, this is the first time the company has confirmed that some features of Apple Intelligence will run on Nvidia chips.
Mike Rockwell, head of Siri engineering at Apple, admitted that the technology was released late. He emphasized that previous plans to revamp Siri did not meet the company's standards.
"We've always had a vision to make Siri the most personalized and powerful virtual assistant available. For the first time, you can ask for information on your device and call the appropriate task. We've optimized the operation so everyone can use it easily," Rockwell shared.
Amar Subramanya, Vice President of AI at Apple, stated that the AFM Cloud Pro has performance comparable to advanced Gemini models. This model will operate on a cloud system using Nvidia GPUs, although it will still be part of Apple's Private Cloud Compute platform.
"We're partnering with both Google and Nvidia to expand our Private Cloud Compute infrastructure, using Nvidia GPUs on Google servers, while still ensuring Apple's superior security policies," CNBC quoted Subramanya as saying.
![]() |
Attendees watch the opening session of WWDC 2026 at Apple headquarters on June 8. Photo: Bloomberg . You may also like |
Using third-party infrastructure presents numerous security challenges. Sebastian Marineau-Mes, Vice President of Software, stated that Apple wants to use the latest chips from Nvidia, while also requiring the chips to be configured in a more private manner. The system must ensure that the servers cannot read user data.
According to Marineau-Mes, recent innovations by Nvidia, such as "ambiguous confidential compute" technology, allow Apple and Google to build systems that meet their own standards.
"We wanted to leverage the latest technology from Nvidia. Therefore, we expanded our Private Cloud Compute system to third-party infrastructure," Marineau-Mes shared.
Apple is differentiating itself from other AI companies by emphasizing privacy. The company doesn't collect data in the way ChatGPT or Claude do; instead, it leverages locally stored information, such as calendars or messages, to personalize AI features.
Google's role
At the press conference, Apple leaders discussed in more detail how Siri AI and Apple Intelligence are being built, while also highlighting the differences from other familiar models.
Federighi stated that the orchestrator between the operating system and software is a crucial element for system security architecture. This component functions to direct AI queries to the appropriate model (running offline or online), based on the computing power and data required.
The relationship between Apple and Google has also been shared in more detail. According to CNBC , Apple Intelligence uses a model developed in-house, not the publicly available Gemini model. The company also does not utilize its partner's existing infrastructure.
![]() |
Improvements to Apple Intelligence were introduced at WWDC 2026. Photo: Apple . |
Specifically, Google's technology was only used to help Apple build its own model. Also known as AFM Generation 3, this model is optimized for Apple Silicon processors.
"The models we just introduced, AFM Core, Core Advanced Cloud, and Cloud Image, are all built specifically for Apple Silicon. They are trained using proprietary data with reinforcement learning methods, and then refined based on results from advanced Gemini models," an Apple representative emphasized.
According to Bloomberg , privacy is a challenge for Apple as the company seeks to offer secure AI features while catching up with its competitors.
"Each model represents a significant leap forward in quality and capability compared to the previous generation. Our goal is to meet all user requirements with the best response time and lowest latency," Subramanya emphasized.
Source: https://znews.vn/apple-xac-nhan-hop-tac-voi-nvidia-post1658162.html












