Apple is reportedly finalizing a $1 billion-a-year deal with Google to use its Gemini AI, a move explicitly described as an “interim solution” as the iPhone maker scrambles to catch up in the generative AI race. This pact will give Apple access to a 1.2 trillion parameter model to power a comprehensive overhaul of Siri, set for release next spring. The deal underscores Apple’s uncharacteristic admission that its in-house technology is not yet ready for prime time, forcing it to rely on a major competitor.
The agreement was reached after Apple benchmarked Google’s Gemini against offerings from OpenAI and Anthropic, concluding that Google’s technology was the most suitable to power the ambitious “Linwood” (the new Siri’s code name) project. This 1.2 trillion parameter model will provide a massive leap in complexity and power from Apple’s current 150-billion parameter models, enabling advanced “summariser” and “planner” functions that are critical for a modern voice assistant.
The project to integrate this third-party AI, “Glenwood,” is being overseen by top Apple executives, including Mike Rockwell and Craig Federighi. While the partnership is substantial, it will be kept “behind-the-scenes.” Apple will not publicly promote Google’s involvement, treating it as a technology supplier, not a partner. This silent arrangement allows Apple to maintain the illusion of its own AI prowess while using Google’s engine under the hood.
A cornerstone of the deal is the protection of user data. The 1.2 trillion parameter Gemini model will be run on Apple’s own Private Cloud Compute servers, not Google’s. This “walled-off” system ensures that Google has zero access to Apple’s user data, a critical requirement for Apple to uphold its privacy-centric brand promise. Apple has already allocated the necessary AI server hardware to manage the significant computational load of the advanced model.
Apple’s ultimate goal is to end this reliance. The company’s management is pushing its internal AI teams, despite recent talent departures, to finish developing a 1 trillion parameter cloud-based model. The hope is to replace Gemini with this in-house solution as early as next year. However, with Google’s Gemini 2.5 Pro already topping AI leaderboards and continuously improving, the “interim” period of this $1 billion deal may last much longer than Apple hopes.
37