Google unveils Private AI Compute for secure cloud-based AI tasks
Google has announced a new AI processing platform, Private AI Compute, designed to deliver advanced artificial intelligence-powered experiences to users while maintaining a strong focus on privacy and data protection. The platform enables AI tasks to be handled in the cloud using Google's latest Gemini models, but with data processing architecture that, according to the company, keeps user information private and inaccessible to anyone else, including Google itself.
Cloud-based processing
The underlying architecture of Private AI Compute uses Google's custom-built Tensor Processing Units (TPUs), which are also deployed in Google's in-house infrastructure for services such as Gmail and Search. The platform pools these resources in a specialised cloud environment called Titanium Intelligence Enclaves, which operates as a secured enclave, isolated from the broader cloud environment.
With this set-up, users can benefit from more powerful AI-driven experiences, such as enhanced suggestions and faster responses, that would be impossible or significantly slower if run solely on the device. According to Google, these improvements are delivered without changing expectations around privacy or shifting additional risk to user data.
Security protocols
Private AI Compute incorporates multiple layers of privacy and security measures. Remote attestation and encryption are used to verify device authenticity and create a communication channel to the cloud-based enclave. Data processed within this environment is isolated from external access, and Google emphasises that even its internal teams cannot view or extract this sensitive information.
The new initiative extends the same privacy expectations associated with on-device processing to the cloud. This is achieved by restricting access at the hardware level and maintaining separate, sealed environments for different user sessions and data sets.
User applications
Initial uses for Private AI Compute are focused on AI features across Google's devices and apps. On Pixel 10 phones, the Magic Cue feature is set to provide more timely and contextually appropriate suggestions by leveraging the new platform's processing power. Additionally, the Recorder app will be able to produce real-time summaries of transcriptions in a wider variety of languages, reflecting a broader capability powered by cloud processing while preserving on-device privacy standards.
Google confirmed that this approach represents a deliberate shift. With coherent integration of local and cloud resources, users can now engage in more complex AI tasks for sensitive applications-such as summarising personal conversations or generating reminders-without compromising data security.
Process transparency
Google stated that development of Private AI Compute has been guided by its previously established AI and privacy principles. The company says it views privacy innovation as a necessary backbone for future AI experiences that are useful and trustworthy to users, especially where personal data is involved.
"Private AI Compute allows you to get faster, more helpful responses, making it easier to find what you need, get smart suggestions and take action," said Jay Yagnik, Vice President, AI Innovation and Research, Google.
The platform is positioned as a response to evolving user expectations and the continued progression of AI. As AI systems take on more proactive and personalised roles-anticipating needs and automating tasks-the need for both computational power and robust privacy technologies has become more pressing.
"Private AI Compute opens up a new set of possibilities for helpful AI experiences now that we can use both on-device and advanced cloud models for the most sensitive use cases," said Yagnik.
