NVIDIA’s confidential computing moves deeper into Apple’s Private Cloud Compute, now running on Google Cloud

At Apple’s WWDC developer conference, NVIDIA said its “confidential computing” technology is being used to support confidential inference for Apple’s Private Cloud Compute (PCC), a system designed to run sensitive parts of Apple Intelligence workloads without exposing user data to the broader computing stack. The update matters because it ties NVIDIA’s security approach to a major privacy architecture Apple is expanding beyond its own data centers, with Google Cloud now included as a deployment venue.

NVIDIA’s announcement describes PCC as expanding from Apple-controlled infrastructure to third-party cloud capacity on Google Cloud. In that broader footprint, NVIDIA said its GPUs with confidential computing capabilities are being used for confidential inference for “Apple Foundation Models,” models described as custom-built by Apple and Google and tied to the technologies behind the Gemini model family.

The companies also framed the rollout as part of their collaboration to support next-generation Apple Intelligence features. NVIDIA said it is working with Apple and Google to integrate its Blackwell GPUs, using confidential computing integrated into PCC’s hardware security architecture, in deployments running on Google Cloud.

Confidential computing is NVIDIA’s term for a hardware-based security layer intended for accelerated AI workloads. NVIDIA said it works by isolating workloads in trusted execution environments, and then using cryptographic verification to help confirm that the underlying infrastructure has not been tampered with before sensitive data is sent to the server.

NVIDIA said the end-user goal is that private inputs, including chats and conversations, cannot be viewed by other parties, “not even the system’s builders.” In practical terms, that claim centers on limiting what the broader cloud provider or infrastructure operator can observe when inference runs in a protected execution environment.

For NVIDIA, the announcement reinforces how data privacy requirements are becoming part of the AI infrastructure conversation, not just model quality or speed. As Apple Intelligence blends on-device processing with cloud-based “server-side inference,” NVIDIA said there is increasing demand for high-performance compute that can still preserve strong privacy and security guarantees.

The sector context is that many AI services, especially those supporting personal assistants and enterprise applications, must handle sensitive information while meeting privacy expectations from users and regulators. PCC’s expansion to Google Cloud suggests Apple is trying to scale compute for intense tasks while keeping its security posture consistent across environments, an approach that is difficult to replicate without hardware-assisted protections.

Still, the details disclosed in NVIDIA’s release are high-level. NVIDIA did not specify model sizes, performance metrics, capacity figures, or which specific Apple Intelligence functions are handled under the confidential inference pipeline. It also did not describe any independent audits or certifications in the statement, nor did it provide implementation depth on how PCC’s hardware security controls interact with Google Cloud’s broader environment.

Going forward, investors and customers will likely watch for additional technical detail on PCC’s confidential inference implementation and for whether Apple expands PCC further across more regions or cloud providers. For NVIDIA, the practical announcement to monitor is whether this kind of confidential computing feature becomes a broader requirement in AI deployments where privacy, governance, and third-party infrastructure use all intersect.