Google presents a new AI cloud environment that is “as secure as local processing”.
Google is once again expanding its AI strategy with Private AI Compute, a new cloud environment that aims to combine powerful generative AI capabilities with the same privacy as local processing. According to the company, devices can now communicate directly with this secure space via an encrypted connection, powered by Google’s own Tensor Processing Units (TPUs).
A Confidential Cloud
Google would have no access to user data. This is because it runs within the Trusted Execution Environment (TEE), which encrypts memory and isolates it from the host system. ArsTechnica reports that an audit by NCC Group confirmed the system meets Google’s strict privacy guidelines.
Until now, Google’s focus has been primarily on on-device AI through the Gemini Nano models on Pixel devices. These process data locally via the NPU and send nothing to the cloud. However, computing power remained limited. Google states that Private AI Compute is “as secure as local processing”. The cloud has much more computing power than a smartphone or laptop, and therefore enough to run the most advanced Gemini models.
Edge versus Cloud
With Private AI Compute, such functions can temporarily and securely send data to the cloud to generate more comprehensive suggestions. For example, Magic Cue in the latest Pixel update will use this environment to provide smarter contextual recommendations, while the Recorder app can now generate summaries in multiple languages.
Google is working toward a standard hybrid model: fast, reliable NPU functions for simple tasks, combined with cloud power for complex AI processing. “Private AI Compute allows us to make generative AI accessible without compromising privacy,” the company says.
There’s a high likelihood that more features from Pixel devices and other Google products will soon switch to Private AI Compute as well.