Apple will integrate Google’s powerful Gemini AI into its new Siri assistant, a deal valued at $1 billion annually for Google. However, the partnership has been meticulously structured to uphold Apple’s stringent privacy commitments.
The core of the agreement lies in the technical implementation. Google’s 1.2 trillion parameter Gemini model will not run on Google’s servers. Instead, it will be hosted on Apple’s own Private Cloud Compute infrastructure, a critical, non-negotiable point for Apple.
This “walled-off” architecture is designed to make it an absolute certainty that Google never gains access to Apple’s user data. This allows Apple to leverage its rival’s superior technology while maintaining its privacy-first marketing and user trust.
This “interim solution” was deemed necessary after Apple’s “Glenwood” project team, overseen by Craig Federighi and Mike Rockwell, found its own 150-billion parameter models insufficient. Google’s AI won a “bake-off” against OpenAI and Anthropic for handling complex “planner” and “summariser” tasks.
The new “Linwood” Siri, set for a spring release, will be a hybrid system. While Apple works to develop its own 1 trillion parameter model to replace Gemini, this “behind-the-scenes” partnership, secured by its privacy guarantees, could last for years.
