Google Cloud says it has built a confidential AI serving platform with Apple that uses Intel TDX, NVIDIA Confidential Computing and Google’s Titan technology to protect data during AI inference. Apple says the work extends Private Cloud Compute to NVIDIA GPUs in Google Cloud for its AFM 3 Cloud Pro model while prese...
Google Cloud said it built a confidential AI serving platform with Apple to protect data-in-use during high-performance AI inference.
In a Google Cloud Blog post titled “Powering the next era of Confidential AI,” Google Cloud said the platform was developed with Apple and combines Intel Trust Domain Extensions, NVIDIA Confidential Computing and Google’s own security technology to support AI inference while protecting data during processing.
The announcement is closely tied to Apple’s Private Cloud Compute, or PCC, an architecture Apple uses for cloud-based processing of some Apple Intelligence requests. In a post from Apple Security Research titled “Expanding Private Cloud Compute,” Apple said PCC on Google Cloud uses NVIDIA Confidential Computing with NVIDIA GPUs, Intel CPUs with TDX and Google’s Titan chip while preserving PCC’s privacy and security goals.
Apple’s Machine Learning Research group also referenced the work in its paper-style post, “Introducing the Third Generation of Apple’s Foundation Models.” Apple said that for AFM 3 Cloud Pro, it worked with Google and NVIDIA to extend Private Cloud Compute to NVIDIA GPUs in Google Cloud while maintaining user-privacy guarantees.
The shared theme across the Google Cloud and Apple posts is protection for data while it is being processed, often called data-in-use protection. Traditional encryption protects data at rest or in transit, but AI inference requires models and user inputs to be processed by hardware. Confidential computing aims to reduce the exposure of that processing by using hardware-backed trusted execution environments and related verification mechanisms.
Google Cloud said Intel TDX is used to help isolate CPU-based workloads, while NVIDIA Confidential Computing extends protection to GPU-accelerated inference. Apple’s Security Research post said the PCC deployment on Google Cloud also uses Google’s Titan chip, which is part of Google’s hardware-rooted security architecture.
The companies are positioning the design as a way to run demanding AI workloads in the cloud while limiting what cloud operators or other parties can access. Apple’s posts specifically frame the work as an extension of PCC’s existing privacy and security model rather than a replacement for it.
Apple said AFM 3 Cloud Pro uses the expanded PCC deployment with NVIDIA GPUs in Google Cloud. That detail is significant because large model inference often depends on GPU acceleration for performance and efficiency. Apple’s Machine Learning Research post states that Apple worked with Google and NVIDIA to bring PCC to NVIDIA GPUs for this model while maintaining user-privacy guarantees.
Google Cloud’s blog similarly emphasizes high-performance AI inference as the use case. The combined approach described by the companies suggests an attempt to reconcile two requirements that are sometimes in tension: running large AI models efficiently and preserving strict privacy boundaries for user requests.
The announcement reflects a broader industry focus on “confidential AI,” a term used for applying confidential computing techniques to AI systems. The Google Cloud post presents the Apple collaboration as part of that direction, while Apple’s own security and machine learning posts give the customer-side context: expanding Private Cloud Compute to support newer cloud-hosted foundation models.
The sources do not provide independent benchmark results in the supplied excerpts, and they do not disclose every operational detail of the deployment. What they do establish is that Apple, Google Cloud, Intel and NVIDIA are involved in a production-oriented effort to support AI inference with hardware-based protections across CPUs, GPUs and platform security components.
For enterprises and consumer technology companies evaluating cloud AI, the practical issue is whether confidential computing can provide verifiable protections without undermining model performance. Google Cloud and Apple’s posts argue that this deployment is a step in that direction, with Apple specifically tying the work to PCC privacy and security goals for AFM 3 Cloud Pro.
Google Cloud said it built a confidential AI serving platform with Apple to protect data in use during high performance AI inference.
The announcement is closely tied to Apple’s Private Cloud Compute, or PCC, an architecture Apple uses for cloud based processing of some Apple Intelligence requests.
What the technology is meant to protect The shared theme across the Google Cloud and Apple posts is protection for data while it is being processed, often called data in use protection.
Continue reading
