Private Cloud Compute (PCC) is Apple’s cloud inference architecture for Apple Intelligence. When a task exceeds the on-device model, the necessary processing can move to Apple servers instead of sending a user’s entire data store to a general AI service.
What problem does it solve?
On-device AI keeps data local, but model size, memory and power are limited. PCC handles tasks that need longer context, more tool calls or more complex reasoning, allowing Apple Intelligence to split work between device and server.
The goal is not to send every AI task to the cloud. The device handles what it can locally, while more complex requests are escalated to PCC. It is a system-managed hybrid AI architecture.
PCC’s privacy design
Apple says data sent to PCC is used only to complete the request and is not stored. The system is designed around end-to-end privacy, with outside researchers able to inspect relevant software and security claims.
No-retention does not remove every AI risk. Users should still consider what an app is asked to process, while developers need data minimization, permissions and error handling. Privacy architecture reduces exposure but does not replace product security.
How can developers use it?
Apple provides PrivateCloudComputeLanguageModel through the Foundation Models framework. Eligible apps can use a Swift API consistent with the on-device model, without managing API keys, server login or a model-provider account.
Apple’s documentation notes daily usage limits. Apps must handle exhausted quotas, unavailable models and network changes, with graceful fallbacks such as returning to the on-device model, shortening the task or asking the user to try later.
The short version
PCC does not move all AI to the cloud. It gives Apple a more private, verifiable and system-integrated path for cloud inference when the device is not enough, especially for apps that need larger models without taking on all the data-governance work of a general AI API.
For official details, see Apple Private Cloud Compute developer page、Foundation Models PCC documentation，以及 WWDC26 developer video。
