Apple PCC goes to Google Cloud; stacks NVIDIA + Intel + Google Titan for confidential inference

Apple announced at WWDC 2026 that it is running Private Cloud Compute (PCC) on Google Cloud for the first time, moving its most privacy-sensitive AI inference workloads off Apple-owned infrastructure. The collaboration uses three-layer hardware trust: NVIDIA Confidential Computing on Blackwell GPUs, Intel TDX on CPUs, and Google Titan as the root of trust — the first time these primitives have been integrated into a global-scale confidential inference pipeline.

Apple's PCC requirements carry unchanged: stateless computation, no privileged runtime access, and verifiable transparency. To minimize trust in Google, Apple maintains its own cryptographically verifiable ledger of all Google Cloud hardware in the PCC fleet rather than relying on Google's attestation. Software attestation is rooted in at least two independent vendors, so compromising a single provider (Intel, NVIDIA, or Google) is insufficient to break verification.

The partnership builds on a multi-year agreement signed in January 2026 for Google's AI models and cloud infrastructure. Practitioners noted the business context: Google built the foundation models powering the next generation of Apple Intelligence, creating a dependency Apple had to engineer around transparently. Google also announced Confidential G4 VMs with NVIDIA Blackwell in preview, lowering the cost of confidential inference beyond H100-based instances.

For teams shipping privacy-critical AI workloads, Apple's approach signals three tiers of cloud inference: Eyes-On (provider sees all), Zero Data Retention (no logs), and Zero Operator Access (cryptographic proof no one, including the provider, can see data). Apple PCC on Google Cloud is ZOA — the strongest tier. The ability to run this on third-party cloud is a significant competitive win for cloud providers willing to invest in confidential computing infrastructure.