A ‘Walled-Off’ AI: How Apple Uses Google’s Gemini Without Sacrificing Privacy

Apple has struck a delicate $1 billion annual deal to use Google’s 1.2 trillion parameter Gemini AI for its new Siri, all while promising to protect user data. The solution is a “walled-off” architecture that is the centerpiece of the agreement.
The “ultrapowerful” Gemini model will be hosted on Apple’s Private Cloud Compute servers, not Google’s. This technical arrangement ensures that while Google’s AI processes the most complex queries, Google itself never gets access to Apple’s user data.
This “interim solution” was necessary for Apple’s “Glenwood” project, the internal plan to fix Siri. After a “bake-off” where Gemini beat OpenAI and Anthropic, Apple chose it to power the “summariser” and “planner” functions in its new “Linwood” assistant.
The technological gap is significant, with Google’s 1.2T model far surpassing Apple’s 150B models. This deal allows Apple to instantly become competitive in AI, a space where it has visibly lagged.
Under the watch of executives Craig Federighi and Mike Rockwell, Apple is still trying to build its own 1T+ model. However, many believe this “behind-the-scenes” partnership with Google could last for years, thanks to the robust privacy system Apple has built around it.