This article may contain affiliate links. Learn more on our Partner Program page.
Apple will now process some of its most demanding AI requests on the infrastructure of its direct competitor. With the expansion of Private Cloud Compute to Google Cloud, Apple's data privacy architecture is leaving its own data centers for the first time – and Apple promises that its data protection guarantees will remain unchanged.
When Apple introduced Private Cloud Compute 2024, the system was the answer to a sensitive question: How to offload AI calculations that are too computationally intensive for the iPhone itself to the cloud without compromising data privacy? Until now, this processing ran exclusively on Apple Silicon servers in Apple's own data centers. With the next generation of Apple Intelligence, this is now fundamentally changing: For particularly computationally intensive tasks, Apple will rely on Google Cloud. This move seamlessly integrates with the previously announced realignment of Apple's Foundation Models to Google's Gemini models and Google's cloud technology, and makes tangible for the first time what this partnership means technically.
What Private Cloud Compute actually does
Private Cloud Compute, or PCC for short, is Apple's cloud system for the private processing of AI requests. Most Apple Intelligence functions run directly on the device, but as soon as a task requires more computing power than is available locally, PCC takes over the calculation in the cloud – according to Apple, without storing or analyzing any data. Since its introduction, this processing has been tied to Apple Silicon servers in Apple's own data centers. With the expansion to Google Cloud, third-party servers are now handling a portion of this load for the first time, and Apple's privacy promise extends into a third-party data center for the first time.
Why Google and NVIDIA are involved
The underlying reason is the collaboration Apple entered into with Google this year: The technology behind Google's Gemini model family forms the foundation for the next generation of Apple Foundation Models, which in turn power Apple's intelligence features. These models extend from the device to the cloud. The most demanding tasks—the independent execution of multi-step commands and complex reasoning, as the revamped Siri in iOS 27 is designed to do - cannot be performed entirely locally. For these cases, Apple, together with Google and NVIDIA, has extended its PCC infrastructure to Google Cloud systems that utilize NVIDIA GPUs. Apple describes this as the first extension of its PCC privacy commitments to data centers outside its own network.
What technology is used on Google's servers
While PCC originally relied solely on Apple Silicon, Google Cloud uses a different hardware foundation. Apple identifies three components: NVIDIA Confidential Computing in conjunction with NVIDIA GPUs, Intel processors with the TDX security extension, and Google's own Titan chip. These components take over the role in Google's data centers that Apple Silicon servers played in Apple's own facilities. Apple emphasizes that the outsourcing was successful without weakening existing security mechanisms – the implementation is new, but the security requirements remain the same.
Five basic rules remain untouched
At its core, PCC is based on five principles that Apple explicitly maintains unchanged in the Google Cloud: stateless processing without persistent data storage, technically enforceable guarantees, no privileged runtime access to the systems, the non-addressability of individual user requests, and verifiable transparency that external security researchers can examine. These five points are the promise by which PCC has always been measured – and according to Apple, they apply equally to third-party hardware as to its own.
Protection beyond Confidential Computing
Apple states that it does not rely solely on its partners' confidential computing technology. Attacks that exploit privileged access outside the confidential computing unit - including side-channel attacks - are additionally protected by treating every layer, from the firmware stack through the host and guest operating systems to the application code, as part of its trusted computing base.
To protect against manipulated hardware in the supply chain, Apple maintains a cryptographically verifiable, exclusively expandable register of all Google Cloud hardware belonging to the PCC fleet. For components that could be used to steal user data in the event of a compromise, software attestation relies on at least two independent trust anchors from different providers.
Architecturally, PCC on Google Cloud also adopts the patterns that have already proven successful on Apple Silicon: The initial reading of incoming network data takes place in a separate process with a segregated namespace, shared inference software is discarded and rebuilt at short intervals, and the attested keys reside in a separate, confidential computing unit isolated from external input.
Control remains with Apple
Regardless of where the infrastructure is located, Apple retains full control over the PCC software. Apple devices trust only PCC software that Apple has cryptographically authorized – a server in a Google data center doesn't change that. The argument is that this ensures user data remains protected, even outside of Apple's own hardware, by the same security and privacy features that have distinguished PCC from the beginning.
Gradual start over the summer
PCC is not yet fully implemented on Google Cloud. Apple is rolling out the complete set of security measures gradually throughout the summer preview phase – in parallel with the beta testing of the next system generation. The details published now therefore describe the target state, not the final version at launch.
Open examination for security researchers
Apple intends to uphold its transparency promise with the new platform: All PCC binaries on Google Cloud will be available for public review. Through the Apple Security Bounty Program, Apple will also provide public research tools and access to active PCC nodes in research mode, enabling external experts to independently verify the data privacy commitments. Apple will announce further technical details at the Confidential Computing Summit later this month; an updated PCC Security Guide and details about the research program are expected later this year.
Apple: Private AI on external servers – now the proof counts.
Data privacy has been Apple's central selling point for years, and running some of its most private calculations on the cloud of a major AI competitor is symbolically problematic. Apple addresses this contradiction not with appeals to trust, but with architecture: verifiable transparency, a cryptographic hardware registry, and the statement that the devices only accept Apple-signed software anyway. Whether this design lives up to its promises in practice will be judged by the very independent security researchers to whom Apple grants access to the live systems. This is precisely the crux of the matter: Apple is making its data privacy model open for public scrutiny not despite, but precisely because of the external location – thus making proof a condition of the entire process. (Image: Apple)
- Siri AI is free – iCloud+ raises AI daily limits
- iOS 27 and more: The new features beyond the WWDC stage
- iOS 27 cleans up the AirPods settings
- iOS 27: Evidence of the foldable iPhone is mounting
- Xcode 27 and new AI frameworks for developers
- tvOS 27: These are the new features for Apple TV
- iOS 27, macOS 27 and more: Apple launches the Developer Betas
- App Store: Apple gives developers new tools
- Apple expands parental controls with new features
- Siri AI in the EU: DMA delays launch on iPhone and iPad
- Apple Intelligence: The next generation is here
- Siri AI: Apple introduces the completely redesigned assistant
- Tim Cook gets in the mood for WWDC with a "Good Morning" video.
- Gemini could become the upper limit for Apple's AI ambitions
- UK is demanding that Apple implement a feature that already exists
- UK threatens tech executives with jail if children can see nude images
- Epic Games counters Apple's move before the Supreme Court
- Apple Arcade: Nine new games in June and July
- App Store: Ecosystem reaches $1.4 trillion
- Apple TV shows the trailer for "Lucky" starring Anya Taylor-Joy
- Apple opens Europe's first Developer Center in Berlin
- Apple discloses revenue figures in India antitrust dispute
