Apple's Gemini-Powered Siri Nears Launch: Inside the $1.5B Partnership

Apple is rebuilding Siri from the ground up. Instead of relying solely on its own AI models, the company chose Google's Gemini and signed a multi-year deal worth approximately $1.5 billion annually. After evaluating OpenAI and Anthropic, Apple's decision to partner with Google represents the biggest pivot in its AI strategy.

Why Google Gemini: The Decision

Apple officially announced on January 12 that it would adopt Google's Gemini model for next-generation Siri. The decision came after evaluating multiple AI companies including OpenAI and Anthropic. It's unusual for Apple to commit to this level of external dependency.

The key factors were scale and infrastructure. The custom model based on Gemini 3 Pro features approximately 1.2 trillion parameters, representing an eightfold improvement over the 150 billion parameter model previously used for Apple Intelligence. Apple adopted it as a 'white label' solution, designed so users see only Siri without any Google branding.

Core Capabilities of the New Siri

The redesigned Siri features three core capabilities.

On-Screen Awareness: Siri can read and understand content currently displayed on screen. If a restaurant appears in Safari, it can make a reservation without copying the name or address. If a flight confirmation email is open, it automatically adds it to the calendar and sets departure reminders.

Multi-Step Actions: A single natural language request can execute up to 10 sequential actions. "Book me on the next available flight to New York, add it to my calendar, and text Sarah my arrival time" runs as a single workflow.

Personal Context: Siri accesses personal data from emails, messages, and calendars to understand user context.

Technical Architecture: Private Cloud Compute

Apple adopted a unique dual-layer architecture for privacy. On-device models (3 to 7 billion parameters) handle roughly 60% of routine tasks, while requests requiring complex reasoning escalate to a 'Deep Intelligence Layer.' This layer runs the custom 1.2 trillion parameter Gemini model on Apple's Private Cloud Compute infrastructure.

Critically, user data never reaches Google's infrastructure. All processing occurs on Apple's own servers, with Google providing only the model technology. Apple says this approach secures both performance and privacy simultaneously.

Potential Delays and Future Outlook

While March was the target launch window, Bloomberg reported on February 11 that internal testing had uncovered issues. Siri sometimes makes errors processing queries or responds too slowly, and was found falling back on ChatGPT for information even when the Gemini-powered system was capable of handling the request.

As a result, some features may be pushed to iOS 26.5 (May) or iOS 27 (September). Expanded personal data access and multi-app actions from single commands are specifically mentioned as potential delays. However, Apple reaffirmed that the redesigned Siri is "still on track to launch in 2026."

Meanwhile, Apple has signaled this partnership may not be permanent. The company is developing its own trillion-parameter model targeting 2027, and the Gemini deal is interpreted as a bridge strategy. A completely new Siri design, codenamed 'Campos,' is expected to be unveiled at WWDC 2026.