Google this week announced Private AI Compute, a new cloud-based system designed to deliver AI capabilities using its Gemini models while maintaining strict data privacy controls, a framework that closely parallels Apple's own Private Cloud Compute technology.
According to Google, the service enables AI tasks that exceed the processing capacity of local hardware to be handled securely in the cloud without exposing personal data to the company or third parties.
For decades, Google has developed privacy-enhancing technologies (PETs) to improve a wide range of AI-related use cases. Today, we're taking the next step in building helpful experiences that keep users safe with Private AI Compute in the cloud, a new AI processing platform that combines our most capable Gemini models from the cloud with the same security and privacy assurances you expect from on-device processing. It's part of our ongoing commitment to deliver AI with safety and responsibility at the core.
Private AI Compute is built using custom Tensor Processing Units (TPUs) with integrated Titanium Intelligence Enclaves (TIE). These hardware-secured enclaves form an isolated, "fortified space" where AI workloads can be processed without direct access to raw user data. Devices connect to the environment using remote attestation and encrypted channels, ensuring that all data transferred remains inaccessible to Google's engineers or infrastructure administrators.
The system will first power new AI experiences on Pixel 10 devices, such as enhancements to Magic Cue, an AI assistant that provides contextually aware suggestions, and an upgraded Recorder app capable of summarizing transcriptions in additional languages. Both rely on Gemini's larger models in the cloud, which require significantly greater computing resources than on-device NPUs can provide.
Private AI Compute is extremely similar to Apple's Private Cloud Compute, which it launched last year. Apple's system supports Apple Intelligence features and uses custom servers containing Apple silicon chips that operate as verifiable, sealed environments for processing AI tasks.
