In a clear sign of the escalating AI arms race, Apple is reportedly finalizing a $1 billion-per-year deal to license Google’s “ultrapowerful” 1.2 trillion parameter Gemini model. This move is a direct response to Apple falling behind its rivals and is aimed at completely overhauling the Siri voice assistant. The sheer scale of the model—1.2 trillion parameters compared to Apple’s current 150-billion parameter cloud AI—highlights the massive computational power now required to compete in generative AI.
This strategic partnership is a pragmatic, “interim” solution for Apple. After an extensive review that included OpenAI’s ChatGPT and Anthropic’s Claude, Apple’s leadership, including executives Craig Federighi and Mike Rockwell, determined that Google’s technology was the only viable option to quickly deliver the “new Siri” (code-named “Linwood”) by its target launch of next spring. This new assistant, planned for iOS 26.4, will use Gemini’s brain for its most complex functions.
The 1.2 trillion parameter model will specifically handle Siri’s “summariser” and “planner” components. This means Siri will finally be able to understand complex, multi-part requests, synthesize information from various sources, and plan the steps to execute the user’s command. This capability, powered by Google’s technology, is central to the “Glenwood” project, the internal code name for the effort to fix Siri’s long-standing shortcomings.
A critical aspect of the deal is Apple’s insistence on data privacy. The Google model will not be accessed via Google’s cloud; instead, it will be hosted and run on Apple’s own Private Cloud Compute servers. This “walled-off” architecture ensures that Google never sees Apple user data, a non-negotiable point for the iPhone maker. Apple has already dedicated significant server hardware to run this computationally intensive model, underscoring its commitment to this privacy-first approach.
Apple does not want to rely on Google long-term. The company’s management is pushing its own AI teams to develop a 1 trillion parameter in-house model with the goal of replacing Gemini as early as next year. However, Apple is chasing a moving target. Google continues to enhance its Gemini platform, with its 2.5 Pro version already dominating AI benchmarks, making it incredibly difficult for Apple’s teams, which have seen recent talent departures, to close the gap.