Apple using Gemini to train smaller AI models
According to The Information, Apple is using a version of Google’s Gemini large language model to train smaller AI models that can run directly on Apple devices through a process called distillation. The approach would allow Apple to offer AI-powered features locally on iPhones, iPads, and Macs without relying entirely on cloud processing.
The report noted that Apple continues prioritising on-device AI as part of its privacy-focused strategy. However, the company is also reportedly exploring acquisitions to strengthen its local AI capabilities. One of the firms Apple has allegedly considered acquiring is Liquid AI, a startup focused on running AI models efficiently on-device.
Cloud processing still expected for complex AI tasks
Despite its local AI efforts, the report suggests that Apple still faces limitations when handling larger AI workloads internally. According to The Information, the full Gemini model contains trillions of parameters and requires significantly more computing power than Apple’s current Private Cloud Compute infrastructure can efficiently handle.
As a result, Apple is reportedly expected to use Google Cloud infrastructure alongside Nvidia AI hardware for certain AI-related requests, particularly those involving a newer version of Siri.
The report claims that some Siri queries may run through Google Cloud using a licensed Gemini model from Google. Apple has also reportedly approved Nvidia’s confidential computing technology for use in order to strengthen security and privacy protections for cloud-based AI processing.
Nvidia’s compute system may help Apple privacy push
According to the report, Nvidia’s confidential computing technology encrypts data and AI models while they are being processed inside graphics processing units (GPUs). Although this reportedly introduces a slight performance slowdown, it could help Apple maintain its privacy commitments while still handling more advanced AI tasks in the cloud.\
The Information added that Apple continues exploring ways to balance cloud-based AI processing with stronger privacy protections, particularly as AI features become more computationally demanding.
Private Cloud Compute branding expected to remain
The report also stated that Apple is likely to continue using its “Private Cloud Compute” branding for future Apple Intelligence features, even if some workloads no longer run exclusively on Apple-owned server infrastructure.
Apple is expected to announce its next set of AI features during WWDC, which begins June 8.
