Google Launches Private AI Compute for Secure Gemini Model Access
Google has unveiled Private AI Compute, a new secure cloud platform designed to let users run advanced Gemini AI models directly on their devices while maintaining robust data privacy.
How Private AI Compute Works
The platform functions similarly to Apple’s Private Cloud Compute, creating what Google describes as “one seamless Google stack.” It is powered by Google’s custom Tensor Processing Units (TPUs).
While AI tools often require more computational power than a device can handle, Private AI Compute processes the same sensitive data that would typically be handled on-device. The key security lies in the TPUs, which utilize an AMD-based Trusted Execution Environment (TEE). This TEE encrypts and isolates memory, preventing access even by Google itself.
Enhanced Features for Pixel Users
According to a company blog post, Private AI Compute will also extend capabilities for Pixel devices. This includes features like Magic Cue, which provides timely suggestions on the latest Pixel 10 phones.
Additionally, the Recorder app on Pixel phones will gain the ability to summarise transcriptions across multiple languages, leveraging the new secure processing power.
