Apple Routes Some Siri Queries to Google Gemini, Adds Nvidia Confidential Compute
Updated
Updated · letsdatascience.com · May 28
Apple Routes Some Siri Queries to Google Gemini, Adds Nvidia Confidential Compute
6 articles · Updated · letsdatascience.com · May 28
Apple will send some new Siri requests to a licensed Gemini model running in Google Cloud, while using a distilled smaller model locally on devices.
Trillions of Gemini parameters exceed Apple's current Private Cloud Compute capacity, pushing more complex queries off-device and making external cloud processing necessary.
Nvidia's confidential compute—recently approved by Apple—encrypts data and models during processing in Google Cloud, trading slightly slower inference for stronger in-use privacy protections.
The hybrid setup sharpens Google Cloud and Gemini's role in mobile AI infrastructure, with WWDC now a likely venue for details on routing, latency, costs and privacy guarantees.
With its AI running on Google's cloud, can Apple's 'Private Cloud Compute' truly guarantee user data privacy?
Is Apple’s billion-dollar AI deal a sign of falling behind or a masterstroke to control the future AI marketplace?