On November 12, 2025, Google announced a new service called Private AI Compute in an official statement. The service aims to provide cloud‑based AI inference and processing with privacy protections approaching "device‑level." Google described it as "Google’s Private AI Compute blends Gemini’s cloud power with encrypted, zero‑access privacy protections," centering on the creation of short‑lived, controlled, end‑to‑end encrypted compute environments in the cloud.
The workflow comprises three steps: customers perform client‑side encryption of input data on their premises or in their enterprise key manager; the encrypted data is sent to Google Cloud’s ephemeral compute enclave for processing; and the processing results remain encrypted in the cloud so that only the party holding the keys can decrypt and read the plaintext output. Google emphasizes this design achieves "zero access," meaning Google cannot access plaintext data or readable copies of model internal activations during processing.
Technically, Private AI Compute leverages Google’s existing confidential computing capabilities and large model hosting platform, integrating these with inference for the Gemini series of models. The service is tailored for privacy‑sensitive industry scenarios including healthcare, finance, and legal advisory, enabling enterprises to run complex models without exposing sensitive information to the cloud provider.
Google said the service supports integration with existing cloud workflows via API. Enterprises can integrate key management into their own KMS or use third‑party HSMs to ensure keys are never uploaded to Google‑controlled environments. Ephemeral compute environments are destroyed after tasks complete, and memory and storage are cleared to reduce the risk of persistent leakage.
On availability, Google indicated the service will initially be offered in preview to Google Cloud enterprise customers and will expand to more regions and additional Gemini model capabilities over time. Pricing and specific SLAs were not disclosed in the announcement; Google said enterprise customers can negotiate compliance and audit terms during contracting to meet regulatory requirements such as GDPR and HIPAA.
Industry observers say Private AI Compute reflects a new approach by cloud providers to AI privacy: returning privacy guarantees traditionally met by local processing to the cloud through encryption and trusted execution environments, balancing compute scale with compliance. For developers, it enables sending sensitive data to the cloud for high‑quality inference without sacrificing model scale or latency; for security teams, it offers an auditable, controllable encrypted workflow.
Summary: The November 12, 2025 release is both a product launch and a technical response to enterprise privacy demands. Private AI Compute ties "zero access" protections to Gemini’s compute, aiming to deliver near‑device privacy guarantees in the cloud and prioritizing customers with strict data confidentiality requirements.