A new report from The Information says Apple iOS 27 could rely on Google’s Gemini models in two different ways: first to help train lighter AI models that run directly on Apple devices, and second to handle a portion of more demanding Siri queries in the cloud. If accurate, the setup would give Apple a way to keep pushing its privacy-first messaging while still expanding AI features that need more computing power than an iPhone or iPad can provide locally.
According to the report, Apple’s AI team has been using a knowledge distillation approach, where a much larger model teaches a smaller one. In this case, Apple is said to be drawing on Gemini as the teacher model so it can build a lighter system for Gemini on-device AI behavior on Apple hardware. The advantage is straightforward: local processing is typically faster, and user data doesn’t have to be sent to the cloud by default.
The report also says Apple has been looking at startups that specialize in shrinking models so they can run efficiently on consumer devices. One name mentioned is Liquid AI, a Cambridge, Massachusetts-based startup focused on compact models designed for local inference. That lines up with Apple’s broader effort to keep as much AI work on-device as possible instead of routing every request to remote servers.
At the same time, the full Gemini model is far too large to run everywhere within Apple’s current in-house private cloud stack. Because of that, the report claims some Siri requests to Google Cloud could appear in iOS 27 when a user asks for something that needs more heavyweight AI processing. Apple would reportedly use an authorized Gemini deployment rather than a consumer-facing chatbot endpoint.
To reduce the privacy risk of that cloud path, Apple is also said to have approved NVIDIA confidential computing technology in recent weeks. The idea is to encrypt data and model handling at the GPU level, which could help protect sensitive requests even though it may slow response times somewhat. If this plan ships, Apple’s future cloud AI stack may depend on both Google Cloud infrastructure and NVIDIA AI hardware.
None of this has been formally announced by Apple, so it still belongs in the leak category for now. Even so, the report offers one of the clearest looks yet at how Apple may balance local privacy, cloud-scale AI, and a more capable Siri experience in Apple iOS 27.
