Apple is extending some of its most demanding Apple Intelligence processing to Google Cloud. PCC has run beyond Apple’s own data centers for the first time.
This changes where some requests run, not the privacy question. Confidential computing means protecting data while it is processed inside hardware-isolated environments.
Apple says the same security rules apply on Google Cloud, but no system can promise zero risk. The key test is whether Apple’s privacy protections remain verifiable on infrastructure operated by another company.
Key Takeaways
- Apple added Google Cloud as a place to run some PCC workloads.
- Apple’s own data centers remain part of the system.
- More demanding AI tasks may need an internet connection.
- Apple retains approval of PCC software.
- Google’s Private AI Compute is a separate service.
How Does Apple Private Cloud Compute Google Cloud 2026 Work?
A request that fits the on-device model can run offline. A harder task can go to PCC. In 2026, Apple added Google Cloud for selected workloads, while Apple data centers continue to operate. The expansion adds capacity; it does not replace Apple’s own servers.
Cloud compute can help with complex reasoning and agentic tool use, where an AI model plans steps or uses tools to answer. The Apple Private Cloud Compute server handles demanding requests, rather than every Apple Intelligence task. More about Apple’s AI processing appears in this guide to Siri and Gemini.
Which Apple Intelligence requests use PCC?
Requests that need more than the on-device Foundation Model can use PCC. Apple’s 2026 developer video lists a 32K context size for the PCC model, compared with 4K for the on-device model. Context size refers to how much text a model can handle at once.
The PCC model offers Light, Moderate, and Deep reasoning levels. The on-device Foundation Model works offline, while a PCC request needs an internet connection. Details on Google Cloud’s AI and compute services help explain its role as an infrastructure provider.
How PCC differs from on-device processing
Apple Foundation Models can run on a device without sending a request to a server. PCC is the cloud option when a task needs more computing power. Google introduced its separate Private AI Compute service on November 11, 2025. That service uses Gemini models, custom Tensor Processing Units, and Titanium Intelligence Enclaves.
Apple and Google therefore offer different private cloud systems. Apple controls PCC software, while Google supplies infrastructure for selected Apple workloads.
A cloud request is not available offline, even when the app can use an on-device model.
Apps may use different processing routes for different features, so a single app does not always send every task to PCC.
What Protects PCC on Google Cloud?
PCC on Google Cloud combines protections from NVIDIA, Intel, and Google with software controls that Apple continues to approve. Confidential computing protects data while it is processed inside hardware-isolated environments. Apple applies five requirements to PCC, including verifiable transparency and limits on privileged runtime access.
The hardware layers support PCC infrastructure, but they do not remove every possible risk. Apple’s software attestation checks the state of software and hardware. Components that could expose user data rely on at least two separate roots of trust from independent vendors.
Hardware roots of trust
The Google Cloud system uses NVIDIA Blackwell GPUs with NVIDIA Confidential Computing, Intel CPUs with TDX, Google’s Titan chip, and a co-engineered open-source host stack. The GPU protects data in use. Intel TDX provides hardware isolation, while Titan supports hardware and boot integrity.
| Component | Technology | Security role |
|---|---|---|
| GPU | NVIDIA Blackwell with NVIDIA Confidential Computing | Protects data in use |
| CPU | Intel TDX | Hardware-based isolation |
| Hardware root of trust | Google Titan | Supports hardware and boot integrity |
Software controls and public verification
Apple’s five PCC requirements are:
- Stateless computation
- Enforceable guarantees
- No privileged runtime access
- Non-targetability
- Verifiable transparency
Apple keeps a cryptographically verifiable, append-only ledger of Google Cloud hardware in the PCC fleet. Initial network data for each request is parsed in a dedicated process with its own namespace. Shared inference software is recycled after a short time-to-live.
Attested keys stay in a separate confidential VM, isolated from external inputs. Apple controls PCC software, and Apple devices trust only software Apple has approved cryptographically. This PCC security overview explains how those controls fit together.
Apple planned to publish PCC binaries and offer research tools and live nodes in research mode through its Apple Security Bounty Program. The privacy controls in AI apps can differ, so users should not assume all services protect data in the same way.
Apple’s PCC GitHub repository includes Cloud Attestation-GoogleCloud, CloudBoard-GoogleCloud, and cloudassetdownloader-GoogleCloud for independent research.
Public binaries and live research-mode nodes give independent researchers a way to examine the system rather than rely only on assurances.
How Can Developers Access PCC, and What Is Apple’s Infrastructure Plan?
Eligible small developers can use Apple Foundation Models on PCC without a cloud API charge if they meet Apple’s program, download, and entitlement conditions. That access is limited to eligible App Store Small Business Program developers with fewer than 2 million first-time App Store downloads across their apps.
Developers also need the PCC entitlement. Users face daily request limits tied to their iCloud accounts, with higher limits available through iCloud+. Apple’s framework supports PCC requests from apps, including watchOS apps.
Developer eligibility, limits and cost
Apple Private Cloud Compute pricing differs for developers and users. Eligible developers pay no cloud API cost under the program, while users have daily limits and can upgrade to iCloud+ for higher limits. Apple has not announced a paid consumer PCC plan or price.
If an app passes 2 million first-time downloads, or its developer leaves the Small Business Program, Apple requires migration within six months. Developers can review Apple’s broader business context in this Apple share price overview.
Apple’s U.S. data-center and server plans
Apple operates its own data centers and planned more U.S. capacity in North Carolina, Iowa, Oregon, Arizona, and Nevada. In 2025, Apple said it would spend more than $500 billion in the United States over four years, including on Apple Intelligence infrastructure and data centers.
Apple said its data centers ran on 100 percent renewable energy when it announced those plans. A 250,000-square-foot Houston server plant was slated to open in 2026 and was expected to create thousands of jobs. Houston-built Apple silicon servers are intended to support Apple Intelligence and PCC.
Google’s estimated $20 billion payment to Apple in 2022 covered default search placement, not PCC collaboration. The estimate came from antitrust court documents and could be higher in later years. More on the Google Cloud data platform shows how broad its cloud business is.
Developers who cross an eligibility threshold need to migrate to another solution within six months.
Conclusion
Apple’s Google Cloud deployment changes where some PCC work runs, not who approves the PCC software. The expansion adds infrastructure while Apple keeps control of software approval.
That distinction matters to users weighing privacy and convenience. Cloud processing can support demanding tasks, while the offline model can handle requests without a connection.
When connectivity matters, check the feature’s processing option before relying on it. If offline access matters, check whether the feature can use the on-device Foundation Model before relying on it.
FAQ
Is private cloud compute really private?
PCC is designed to limit access to user data, but no computer system is risk-free. It uses stateless processing, restricts privileged runtime access, and supports public checks of its software and hardware state. Apple also plans research access to PCC binaries and live nodes. These measures let researchers test claims, but they cannot promise that every risk has disappeared.
Is Google paying $20 billion to Apple?
No. The estimated $20 billion payment was for Google’s default search placement on Apple devices in 2022, not for PCC or Google Cloud work. The estimate came from antitrust court documents and could be higher in later years. The search deal and PCC infrastructure collaboration are separate arrangements, so the figure does not describe the cost of Apple’s cloud processing.
How does Apple Private Cloud Compute work?
An Apple device can send a demanding AI request to a PCC node when its on-device model is not enough. PCC processes the request in a protected environment, while Apple controls which software its devices trust. The request needs an internet connection. Less demanding tasks may stay on the device, where the on-device Foundation Model can work offline.
What is the point of a private cloud?
A private cloud lets a company run more demanding AI inference remotely while adding controls meant to limit access to user data. That extra computing power can help with complex tasks that a phone model cannot handle as well. In Apple’s case, PCC sends selected Apple Intelligence requests to cloud servers. Users still need an internet connection for those requests.
How to access Apple Private Cloud Compute?
Users access PCC through supported Apple Intelligence features, rather than by choosing a separate cloud app. Developers can use the Foundation Models framework to add Apple model features to apps. The no-cloud-API-cost program has added conditions: developers need Small Business Program eligibility, the required download level, and a PCC entitlement. User requests also count against daily iCloud limits.
Does Apple have cloud computing?
Yes. Apple operates its own data centers and expanded PCC to Google Cloud for selected workloads. The move gives Apple another place to run some demanding AI tasks, while Apple’s own facilities remain in use. Google provides infrastructure for this PCC work, but Google’s separate Private AI Compute service is not the same service as Apple PCC.
Is Apple Intelligence safe to use?
Apple Intelligence has layered safeguards, but no service can promise zero risk. PCC uses hardware isolation, software checks, and public research access to support its privacy claims. Users should still consider what an app feature sends to the cloud and whether it needs an internet connection. A task that can run on-device may be a better fit when offline use matters.
Does Apple have its own data center?
Yes. Apple operates its own data centers and planned additional U.S. capacity in North Carolina, Iowa, Oregon, Arizona, and Nevada. Apple also planned a Houston facility to assemble Apple silicon servers for Apple Intelligence and PCC. The Google Cloud expansion adds another infrastructure option; it does not mean Apple has stopped using its own data centers.
