In a deal valued at $1 billion annually, Apple is set to integrate Google’s 1.2 trillion parameter Gemini AI model to serve as the new “summariser” and “planner” for its redesigned Siri. This move signifies that Google’s technology will handle the most cognitively demanding tasks for Apple’s assistant, such as synthesizing large amounts of information and planning the execution of complex, multi-part commands. This is a core component of Apple’s “Glenwood” project to fix the assistant, with the new “Linwood” version of Siri slated for release next spring.
The decision to outsource Siri’s “brain” to a rival comes after Apple’s own models were deemed insufficient for the task. Apple tested Gemini against OpenAI’s ChatGPT and Anthropic’s Claude before zeroing in on Google’s offering as the most powerful interim solution. The 1.2 trillion parameter model is a monumental upgrade from Apple’s current 150-billion parameter cloud AI, providing the necessary power to process complex data and understand user context at a much deeper level.
This “behind-the-scenes” partnership, led by Apple executives Craig Federighi and Mike Rockwell, will not be publicly promoted. Apple’s strategy is to use Google as a silent technology supplier, allowing the new Siri to be presented as an Apple innovation. This is a different dynamic than the public-facing Safari search deal, where Google’s brand is visible. This new pact is also separate from earlier, unsuccessful talks to add Gemini as a standalone chatbot.
Privacy is a central pillar of the agreement. To prevent Google from accessing any user data, the massive Gemini model will be hosted and run on Apple’s proprietary Private Cloud Compute servers. Apple has already allocated the server hardware required for this computationally intensive task. This “walled-off” approach allows Apple to leverage Google’s superior AI model without violating its core privacy principles, a critical balance for the company.
Apple’s reliance on Google is not intended to be permanent. The company’s management is still pushing its internal AI teams to develop a replacement, reportedly a 1 trillion parameter cloud-based model, with the hope of launching it as early as next year. However, Apple faces a significant challenge in catching up to Google, whose Gemini 2.5 Pro model continues to set the standard in AI performance benchmarks, making this “temporary fix” a difficult one to escape.