Apple’s AI ‘Catch-Up’ Plan: Pay Google $1B/Year for 1.2T Parameter Gemini

Apple has reportedly devised a $1 billion-a-year plan to catch up in the generative AI race: pay Google to use its technology. The iPhone maker is finalizing a deal to license an “ultrapowerful” 1.2 trillion parameter version of Google’s Gemini AI. This model will serve as the engine for a long-overdue overhaul of Siri, rebuilding the assistant’s underlying technology to power a new slate of features set to launch next year.
This strategic partnership is a clear admission that Apple’s in-house AI development has fallen behind. After an extensive review of third-party models, including OpenAI’s ChatGPT and Anthropic’s Claude, Apple’s leadership—including executives Craig Federighi and Mike Rockwell—selected Gemini as the best “interim solution” to bridge the gap. The new Siri, code-named “Linwood,” is on track for a spring release, thanks to this infusion of Google’s technology.
The 1.2 trillion parameter model is a game-changer for Siri. It completely dwarfs Apple’s current 150-billion parameter cloud AI, providing the computational power needed for advanced “summariser” and “planner” functions. These components are essential for enabling Siri to understand complex, multi-part requests and plan the necessary steps to execute them, a key weakness of the current assistant.
To protect user privacy, Apple has structured the deal to keep Google’s hands off its data. The Gemini model will be deployed on Apple’s own Private Cloud Compute servers, which have already been allocated for the task. This “walled-off” infrastructure ensures that all user data remains within Apple’s ecosystem, completely shielded from Google, a non-negotiable term for the privacy-focused company.
While the partnership is substantial, Apple will not be promoting it. Google will act as a “behind-the-scenes” technology supplier, a very different arrangement from the public Safari search deal. Apple’s management still intends for this to be a temporary fix, as its own AI teams are rushing to finish a 1 trillion parameter model. However, catching Google’s top-tier, rapidly evolving Gemini AI remains a formidable challenge.