Inside Apple's Siri Overhaul: The Google Gemini and Nvidia Partnership

How the On-Device and Cloud Split Works

Not every Siri query will require a trip to Google's data centers. Apple is using a technique called model distillation to train smaller versions of the full Gemini model that can run locally on iPhones, iPads, and Macs . For simple, fast tasks — setting timers, sending messages, basic knowledge lookups — the on-device model handles the request entirely on Apple Silicon.

Complex queries are a different story. When a user asks something that requires deeper reasoning, personal context, on-screen awareness, or multi-step actions across apps, the request is routed to Google Cloud . There, the full 1.2 trillion-parameter Gemini model takes over, running on Google's fleet of Nvidia Blackwell B200 GPUs — Nvidia's flagship data-center chip designed for trillion-parameter AI workloads .

This is a significant pivot. Apple had previously asserted that Apple Intelligence would run exclusively on Apple Silicon . The Blackwell B200 marks the company's first major reliance on non-Apple hardware for its AI assistant .

Nvidia's Role and Confidential Computing

The Nvidia angle, first reported in detail by The Information in June 2026, adds a critical layer to the story. According to the report, Apple recently approved the use of Nvidia's confidential computing technology, a hardware-level security feature that encrypts data while it is being actively processed on the GPU .

In practice, this means:

• User queries are anonymized before they leave the device. Apple does not share Apple IDs or user identities with Google .
• Data is encrypted in transit and remains encrypted even while the Nvidia GPUs are computing the response .
• Neither Google nor Nvidia can see the plaintext content of user requests .

Apple is framing this as a "privacy-first" architecture . However, some privacy advocates have raised concerns given the earlier promise that Apple Intelligence would stay within Apple's own silicon ecosystem .

Timeline: From Announcement to Launch

The path to the new Siri has moved quickly by Apple's standards:

• January 12, 2026 — Apple and Google formally announce the Gemini partnership via a joint statement .
• March 2026 — Reports emerge that Apple can now distill the full Gemini model into smaller on-device versions .
• WWDC 2026 (June) — Apple is expected to preview the revamped Siri, iOS 27, and the developer betas . The new assistant is expected to demonstrate on-screen awareness, personal context understanding, and the ability to take actions inside and across third-party apps .
• September 2026 — The overhauled Siri launches publicly alongside iOS 27 and the iPhone 18 lineup .

What We Know — and What's Still Unconfirmed

It is worth noting a clear gap between what's officially confirmed and what comes from press reporting. Apple and Google's joint statement confirmed the Gemini-based partnership and a 2026 launch timeline, but it disclosed no details about Nvidia hardware, the specific model size, or the $1 billion licensing fee .

Those specifics — the Nvidia Blackwell B200 chips, the 1.2 trillion-parameter count, and the reported price tag — trace back primarily to a June 2026 report by The Information and subsequent secondary coverage . The September launch window is well-sourced from the same reporting but has not yet been formally announced by Apple.

For now, what is certain is that Apple has bet its most important consumer AI product on Google's foundation models — and that Nvidia's silicon is doing at least some of the heavy lifting behind the scenes.