How Apple Will Use Google’s $1B AI While Keeping Data ‘Walled Off’

Apple is executing a high-stakes balancing act: it is preparing to pay Google $1 billion a year for its 1.2 trillion parameter Gemini AI to fix Siri, while simultaneously guaranteeing that Google will never see any user data. The deal hinges on Apple’s “Private Cloud Compute” infrastructure. The massive Google-built model will be hosted and run entirely on Apple’s own servers, creating a “walled-off” environment that leverages Google’s AI brain without compromising Apple’s privacy-first principles.
This complex arrangement is Apple’s “interim solution” to its generative AI problem. The company’s internal “Glenwood” project, led by Craig Federighi and Mike Rockwell, tested models from OpenAI and Anthropic before concluding that Google’s “ultrapowerful” Gemini AI was necessary to power the new Siri (code-named “Linwood”). This new assistant, set for a spring release, will use Gemini’s power for its “summariser” and “planner” functions, which are critical for handling complex user requests.
The 1.2 trillion parameter model provides a massive upgrade in computational power compared to Apple’s current 150-billion parameter cloud AI. This raw power is needed to process the nuanced, multi-step commands that modern AI assistants are expected to handle. While Gemini will tackle this heavy lifting, some simpler Siri functions will continue to run on Apple’s in-house models.
This partnership is purely technological and will not be marketed to the public. Apple is treating Google as a behind-the-scenes supplier, a stark contrast to the visible Google search deal in Safari. This is also separate from earlier, failed talks about a direct Gemini chatbot integration. Apple’s CEO has previously stated that Siri could eventually offer additional chatbots, but this deal embeds Google’s tech at a much deeper, core level.
Apple’s long-term plan is to escape this reliance. Its own AI teams are working to finish a 1 trillion parameter model to replace Gemini, potentially as soonin as next year. However, with Google’s Gemini 2.5 Pro dominating AI benchmarks, Apple’s teams are under immense pressure to create a model that can not only catch up but surpass a rapidly advancing competitor, all while the $1 billion annual bill continues.