What you need to know
- Apple’s upgraded Siri and Apple Intelligence features are indirectly powered by Google’s Gemini technology.
- Apple says all AI requests run on-device or through Private Cloud Compute without sharing user data.
- Apple Foundation Models are trained by Apple but refined using techniques from Gemini frontier models.
Apple’s upgraded Siri AI stole the spotlight at WWDC today, and it turns out Google Gemini is playing a bigger role behind the scenes than expected.
Earlier this year, Apple announced that it was partnering with Google to leverage Gemini’s capabilities, and today we finally saw the first major result of that collaboration. Apple introduced a number of new Apple Intelligence features across its ecosystem, including a much more capable Siri, improved dictation, and even small and thoughtful features like the ability to create Shortcuts using natural language prompts.
On its website, Apple says these new experiences are powered by its own Apple Foundation Models. However, those models were developed in collaboration with Google and Gemini.
Apple also emphasizes that the models run on-device whenever possible, and when cloud processing is required, requests are handled through the company’s Private Cloud Compute infrastructure. According to Apple, that means user data isn’t shared with either Google or Apple itself.
Apple later clarified that iOS 27 doesn’t directly use Gemini apps or Google’s client-side code to power these features (via 9to5Mac).
That said, Apple’s AI chief, Subramanya, explained that the new Apple Intelligence experience is built on four Apple Foundation Models, or AFMs. At the base level is AFM Core, which is somewhat similar to Gemini Nano and handles on-device AI tasks.
Above that sits AFM Core Advanced, which adds multimodal capabilities. On the cloud side, Apple uses AFM Cloud and AFM Cloud Image models for more demanding requests.
Apple says these models are trained on the company’s own proprietary datasets but are refined using techniques derived from Gemini’s frontier models.
All things considered, while Apple is understandably positioning Apple Intelligence as its own AI tech, it’s pretty clear that Google’s Gemini models played a significant role in helping Apple get there.
How good the actual implementation ends up being is a completely different question, though. We’ll have to spend more time with iOS 27 before making that call.