Apple Nears $1 Billion Deal with Google for Custom Gemini Model to Power Siri

Bloomberg reports that Apple is finalizing a major partnership with Google to use a custom version of the Gemini AI model for the next-generation Siri, set to launch next year. The agreement could be worth around $1 billion annually, marking a major step in Apple’s AI ambitions.

A 1.2 Trillion-Parameter Model

According to Bloomberg’s Mark Gurman, Apple plans to use a 1.2 trillion-parameter Gemini model to power the revamped Siri — a huge leap from Apple’s current 150 billion-parameter cloud model and 3 billion-parameter on-device model.

The collaboration, known internally as Project Glenwood, is led by Mike Rockwell, who recently took charge of Siri’s overhaul.

What Gemini Will Do

Under the deal, Google’s Gemini model will handle Siri’s summarizer and planner functions — the systems that help the voice assistant synthesize information and execute complex tasks. Other Siri features will continue to rely on Apple’s in-house AI models.

Importantly, user data will stay private. Gemini will run on Apple’s Private Cloud Compute servers, ensuring that information never leaves Apple’s infrastructure.

Why Apple Chose Google

Gurman reports that Apple’s decision to go with Google over Anthropic came down not just to performance, but also pricing. After months of evaluation, Google’s offer reportedly proved more cost-effective, leading to the near-finalized $1 billion-per-year deal.

Apple’s Long-Term AI Vision

Despite the partnership, Apple still aims to develop its own large-scale model. The company’s internal team is reportedly building a 1 trillion-parameter model designed to eventually replace Gemini and bring more AI functions fully in-house — potentially as soon as next year.

However, with Gemini powering key parts of Siri’s upgrade, Apple is ensuring that its voice assistant finally gets the AI intelligence it has long been missing.