Apple is preparing the biggest overhaul of Siri in years for iOS 27, according to The Information, and the technical foundation of this update is more complex than one might expect from a company known for its focus on privacy. Journalist Aaron Tillich reported that Apple is using Google Gemini, Google Cloud, and Nvidia chips for the new voice assistant architecture.
This doesn't mean a full Gemini model will run directly on the iPhone. Such AI systems require their own data centers, so Apple is reportedly using Gemini to train smaller local models via distillation. These models can handle everyday Siri commands directly on the device: setting timers, editing text, answering simple questions, and processing basic requests without calling out to the cloud.
More complex tasks that the local model can't handle will be redirected to Google Cloud, where a licensed version of Gemini runs. This complicates Apple's usual narrative about complete on-device data processing, but the company is trying to preserve privacy through Nvidia's confidential computing technology. That approach encrypts data and the model during GPU processing, preventing the cloud infrastructure from accessing identifiable user information.
According to the publication, Google won't be able to use the iOS request stream to train its own models. Still, the trade-off is clear: Apple is temporarily relying on external infrastructure because its own Private Cloud Compute based on M-series chips doesn't yet have enough capacity to handle complex requests at the level of modern AI models.
Users will likely see not the technical side, but the new Siri interface and expanded Apple Intelligence features. Bloomberg previously reported that the updated assistant will debut at WWDC on June 8, along with a new visual design and deeper system integration. The full rollout of features is expected in the fall with iOS 27.
For Apple, this could be a transitional phase. The company is also reportedly considering buying small AI startups to accelerate development of its own compression and on-device model technologies. If these plans pan out, dependence on Gemini and Google Cloud might prove to be a temporary bridge—a way to quickly modernize Siri now while Apple builds its own long-term AI platform.