Apple approves Nvidia chips in Google Cloud for new Siri

Apple Turns to Nvidia Chips in Google Cloud to Power Next-Generation Siri

Apple is using Nvidia confidential computing technology within Google Cloud to process complex AI queries for its upcoming Siri overhaul, even as it plans to retain the “Private Cloud Compute” branding for the service, according to a report from The Information published Wednesday.

The report reveals that Apple is distilling Google’s large Gemini model — training a smaller version capable of running locally on iPhones and other Apple hardware — while routing heavier queries through Google Cloud infrastructure powered by Nvidia GPUs. Apple approved the use of Nvidia’s confidential compute feature “in recent weeks,” according to sources cited by The Information.letsdatascience

Encryption as a Privacy Bridge

Nvidia’s confidential compute is a security feature embedded in its GPUs that encrypts data and AI models while they are being processed, ensuring that neither Google nor other third parties can access user information during inference. The technology adds a modest performance cost but provides stronger protections for data in use, allowing Apple to uphold its privacy commitments even as queries pass through third-party infrastructure.firstpost

The arrangement marks a notable evolution from Apple’s original Private Cloud Compute system, which ran exclusively on Apple-designed silicon in Apple-controlled servers. Despite this architectural shift to Google’s cloud, Apple is expected to continue using the “Private Cloud Compute” branding for its next wave of Apple Intelligence features, people familiar with the partnership told The Information.macrumors

Distillation Strategy Ahead of WWDC

Apple’s broader strategy involves using a large version of Google’s Gemini model as a “teacher” to train smaller “student” models through distillation — a process that compresses the reasoning capabilities of a massive model into one compact enough to run on-device. This approach lets Apple deliver advanced AI features locally on its custom silicon while reserving cloud processing for the most demanding tasks.letsdatascience

The revelations come as Apple prepares to showcase its on-device AI capabilities at WWDC next month, where the company is expected to emphasize how 15 years of custom chip development gives it an edge in running AI models directly on devices. Apple will reportedly position local inference as both a privacy-preserving and cost-saving alternative to the data center expansions pursued by rivals.firstpost

The multi-year Apple-Google partnership, announced in January 2026, established Gemini as the foundation for Apple’s next-generation models, with Apple paying an estimated $1 billion annually for access to the technology.blog