Apple has a secret project, code-named “Glenwood,” with a singular goal: fix Siri. And the price tag for this fix is reportedly $1 billion a year, paid directly to its rival, Google. Apple is finalizing a deal to license Google’s “ultrapowerful” 1.2 trillion parameter Gemini AI model, which will serve as the new engine for a completely redesigned Siri. This “behind-the-scenes” partnership is Apple’s interim plan to deliver a competitive smart assistant by next spring.
The “Glenwood” project is being led by some of Apple’s heaviest hitters, including software engineering chief Craig Federighi and Vision Pro developer Mike Rockwell. Their decision to use Gemini came after a comprehensive review of all major third-party models, including those from OpenAI and Anthropic. Google’s AI was chosen to power the new Siri, code-named “Linwood,” which is expected to launch with iOS 26.4.
This custom 1.2 trillion parameter Gemini system is a massive technological leap. For comparison, Apple’s current cloud AI model is believed to be around 150 billion parameters. This new, more complex model will vastly expand Siri’s ability to process complex data and understand context, allowing it to handle “summariser” and “planner” functions—components that help Siri synthesize information and decide how to execute complex tasks.
A non-negotiable condition from Apple is that user privacy must be protected. To achieve this, the Google-built AI model will run on Apple’s own Private Cloud Compute servers. This “walled-off” infrastructure guarantees that Google will have no access to any Apple user data, allowing Apple to leverage its rival’s technology without compromising its core privacy values. Apple has already allocated the necessary server hardware for this massive undertaking.
Apple’s management still intends to replace Gemini with an in-house solution and views this $1 billion-a-year deal as a temporary fix. Its own models team is working on a 1 trillion parameter cloud-based model it hopes to have ready as early as next year. However, with Google’s Gemini 2.5 Pro already sitting atop AI leaderboards, Apple’s teams are in a difficult race to catch a rapidly accelerating target.
12