Apple Extends Private Cloud Compute to Google Cloud for AI Processing

Apple is significantly expanding the reach of its Private Cloud Compute (PCC) platform, the foundational infrastructure for its upcoming Apple Intelligence features. In a move that broadens its AI processing capabilities beyond its own data centers, Apple will now leverage Google Cloud's infrastructure, marking a substantial shift in how its AI services are delivered.

PCC, first introduced in 2024, is designed to handle AI workloads that exceed the processing power of on-device hardware. Crucially, it aims to maintain Apple's stringent privacy and security standards even when processing data in the cloud. The platform was initially built on Apple silicon and operated exclusively within Apple's controlled environments.

This expansion into third-party data centers, specifically Google Cloud, is underpinned by a commitment to confidential computing technologies. Apple has collaborated with Google and NVIDIA to ensure that PCC's core security model remains intact. This model emphasizes stateless computation, enforceable guarantees, no privileged runtime access, non-targetability, and verifiable transparency, ensuring that user data remains protected.

The security architecture for PCC on Google Cloud incorporates several advanced features. These include the use of NVIDIA Confidential Computing, Intel CPUs with TDX, NVIDIA GPUs, and Google's own Titan chip. Apple is also implementing measures to reduce supply chain risks by maintaining a cryptographically verifiable, append-only ledger of all Google Cloud hardware used in the PCC fleet. For software attestation, components critical for preventing data exfiltration rely on multiple roots of trust from independent vendors.

Further enhancing security, PCC on Google Cloud will mirror existing mechanisms from its Apple-based counterpart. This includes initial network data parsing within a dedicated, isolated process, recycling shared inference software with a short lifespan, and storing attested keys in a separate confidential virtual machine. Apple asserts that these measures will ensure user data is protected by PCC's security properties, even when processed outside of Apple's direct infrastructure.

Apple retains full control over the PCC software stack, irrespective of its deployment location. Apple devices will only trust software that has been cryptographically approved by the company, providing an additional layer of security. The company plans to publish all PCC binaries for public inspection and will offer research tools and access to live PCC nodes in research mode through its Apple Security Bounty program.

This strategic partnership with Google Cloud allows Apple to scale its AI capabilities more effectively while adhering to its long-standing privacy commitments. The gradual rollout of full protections for PCC on Google Cloud is expected throughout the summer preview period, signaling a new era for AI processing in the tech industry.