Apple is inking a deal with Google to make a custom Gemini AI model to power the next version of its virtual assistant Siri for spring 2026, according to a Nov. 2 report by Bloomberg's Mark Gurman. A subsequent report on Wednesday states Apple will pay $1 billion a year for this new 1.2 trillion parameter model.
Apple reportedly was evaluating whether to use Google or AI competitor Anthropic for the next version of Siri. While Apple will pay Google $1 billion a year, it would have cost $1.5 billion per year with Anthropic, according to another Bloomberg report.
Don't miss any of our unbiased tech content and lab-based reviews. Add CNET as a preferred Google source.
This custom Gemini model will run on Apple's private cloud computer servers. Apple's own models will continue to run on devices for personal data, while Gemini will operate on servers for more complex tasks. Apple doesn't plan to highlight Google's involvement in the company's marketing, Bloomberg reports.
Google declined to comment. Apple and Anthropic didn't respond to requests for comment.
Why Apple doesn't build its own AI
With major tech companies pivoting toward AI, Apple has largely been left behind. The tech company was slow to adopt the technology and hasn't been able to develop competitive AI models. It instead turned to companies like OpenAI, the creator of ChatGPT, to help add generative capabilities on top of existing Apple systems.
(Disclosure: Ziff Davis, CNET's parent company, in April filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)
Even so, the promise of a truly agentic Apple Intelligence has failed to materialize, although it has improved. Apple CEO Tim Cook also hasn't ruled out the possibility of acquiring an AI company.
As to why Apple hasn't made significant strides in AI despite being one of the richest companies in the world has likely to do with prioritization. Apple, as a company, focuses on privacy for its customers. Given how data intensive AI systems rely on cloud servers, Apple has been more keen on local AI models, ones that can run on a device. The problem is that AI is computationally intensive and smaller local models simply can't compete with server stacks. Training foundational models also requires a lot of expertise and expenditure. While OpenAI hasn't given exact numbers, estimates put the training cost of GPT-5 to be over $1 billion. AI researchers also command high salaries and like to openly publish their research. For Apple, partnering with existing AI companies to integrate their technology better aligns with its culture.
... continue reading