In a move that highlights its urgent need to catch up in the AI race, Apple is reportedly forming a reluctant alliance with Google, agreeing to pay around $1 billion annually for an “interim” fix for Siri. This deal will give Apple access to a custom 1.2 trillion parameter Gemini AI model, which it will use to rebuild its voice assistant’s core technology. The new Siri, planned for a spring release, will leverage this “ultrapowerful” model to deliver a new slate of features Apple’s own AI cannot yet support.
The decision to partner with a chief rival comes after Apple’s “Glenwood” project team, led by software chief Craig Federighi, benchmarked all top AI models. After testing ChatGPT and Claude, Apple zeroed in on Google’s Gemini as the most capable short-term solution. The goal is to use this technology to power the new “Linwood” version of Siri (planned for iOS 26.4) while Apple’s own 1 trillion parameter model is completed.
The 1.2 trillion parameter Google model will be deployed for Siri’s most complex cognitive tasks: its “summariser” and “planner” functions. This will allow the assistant to understand and plan the execution of multi-step, context-aware commands, a massive leap from its current capabilities, which are based on much smaller 150-billion parameter models. Other Apple Intelligence features will continue to use Apple’s proprietary technology.
A critical component of this reluctant alliance is privacy. Apple has mandated that the Gemini model will be hosted on its own Private Cloud Compute servers. This “walled-off” system, for which Apple has already allocated hardware, ensures that no user data is ever processed by or visible to Google, allowing Apple to use its rival’s AI without sacrificing its core values.
This $1 billion-a-year pact will not be publicly advertised. Apple will treat Google as a silent supplier, unlike the public-facing Safari search deal. This reflects Apple’s desire to eventually replace Google’s technology. However, with Google’s Gemini 2.5 Pro already topping AI leaderboards and improving rapidly, this “interim” solution may prove to be a very costly and long-term dependency for the iPhone maker.