What remains for Apple? Siri will run on Nvidia GPUs with Google’s code

The WWDC 2026 is imminent, and the anticipation for the unveiling of iOS 27 and the highly anticipated Siri 2.0 has reached its peak.

However, behind the scenes, Cupertino’s strategy has undergone a drastic deviation from the initial premises. At the launch of Apple Intelligence, the company had generated significant fanfare by focusing on a closed and secure infrastructure: servers based on proprietary chips and an AI entirely built in-house, designed to guarantee maximum user privacy.

Today, the narrative clashes with a very different reality. The need for computing power and advanced language models has pushed Tim Cook’s company to rely on external giants, effectively forfeiting total control over hardware and software.

Apple: Deals with Nvidia and Google for the New AI

According to recent rumors, the most complex requests directed at the virtual assistant will not be processed by algorithms developed in Cupertino. On the contrary, Apple would have signed an agreement worth 1 billion dollars per year to leverage a customized version of Google’s Gemini model.

This step is impressive: the voice assistant will abandon the current 3-billion-parameter generative model, today used for textual summaries and basic responses, to rely on a colossal architecture of 1.2 trillion parameters. A move to provide highly articulated responses without having to always delegate the most difficult questions to external services like ChatGPT.

I super chip di Nvidia a tutela dei dati

The change affects not only the software infrastructure, but also extends deeply to the processors. To run the new Gemini model within Google Cloud, Nvidia technologies will be employed. Specifically, Apple will leverage the data center chips Blackwell B200.

The choice is driven by the desire to uphold the promise of data security. Nvidia processors, in fact, integrate a confidential computing function that encrypts information directly during processing on the chip, preserving the user’s privacy.

From an engineering standpoint, the B200 offers astonishing performance thanks to parallel processing that leverages thousands of cores to perform operations simultaneously that traditional CPUs would handle sequentially.

The processor is built with technology 4 nm from TSMC and unites two dies into a single component capable of communicating at a speed of 10 TB/s. The result is a chip with 208 billion transistors.

L’assistente personale in azione

Despite the unusual loss of independence, the benefits for the end user will be tangible. The new features will be accessible only on the latest high-end smartphones, starting with the iPhone 15 Pro, continuing through the iPhone 16 series and iPhone 17, up to the future models of the iPhone 18 lineup.

On compatible devices, the new Siri will become a deeply interconnected entity with the operating system. It will have the ability to analyze native applications autonomously such as Messages, Mail, Notes, Photos and Calendar to extract contextual information.

For example, by cross-referencing the data of an incoming flight found in the email inbox with the flight-tracking systems and the traffic information from Apple Maps, the assistant will be able to autonomously calculate the exact time at which the user will have to leave home to pick up a relative at the airport.

A remarkable level of automation that, although it is powered by third-party technologies, will strive to maintain the typical user experience of Apple’s ecosystem.