Introduction

On June 10, 2025, Apple used its WWDC keynote to introduce Private Cloud Compute (PCC). This new architecture brings large‑language‑model horsepower to Apple devices without sending user data to external servers. PCC pairs on‑device Neural Engine upgrades with an encrypted, Apple‑owned micro‑data‑center fabric that processes only anonymized model weights, never raw user prompts.

“We designed AI to be private by architecture, not by policy,” said Apple SVP of Software Engineering Craig Federighi.¹

Here’s how PCC works: When a user asks Siri, or any PCC‑enabled app, a request that exceeds local compute, the device spins up a one‑time encrypted session to an Apple PCC node. The node runs on M4 silicon, boots from a read‑only image, generates an ephemeral key pair, and destroys logs after computation. Third‑party security groups can audit both hardware and firmware at Apple‑owned facilities.

Apple claims this approach combines the latency of on‑device AI with the power of large server models, all while meeting GDPR and forthcoming U.S. AI Accountability Act guidelines.

Why it matters now

• Privacy expectations around GenAI are rising; enterprises need verifiable guarantees.
• Hybrid edge‑plus‑micro‑cloud may beat pure on‑device models in capability.
• Apple’s move forces competitors to justify cloud collection practices.

Call‑out: AI without surveillance—finally in your pocket

Early developer tests show Siri latency dropping from 280 ms to 85 ms, while 92% less user data leaves the device compared to traditional cloud LLM calls.

Business implications

• Regulated industries (health, finance, government) can adopt AI assistants on iPhone and Mac without breaching data‑residency rules.
• Enterprise app developers can target PCC via the new ‘SecureCompute’ API, offloading heavy reasoning while preserving user trust.
• Security teams benefit from deterministic, audit‑friendly AI sessions, each request is cryptographically signed and locally attestable.

Apple is rolling PCC into iOS 19, macOS 15, and visionOS 2 this fall. A beta SDK, available today, allows apps to declare whether prompts may leave the device; User ConsentKit surfaces a toggle per app.

Looking ahead

Apple plans to expand PCC nodes to regional data centers on five continents by 2026, ensuring a sub-100 ms round-trip for 90% of users. The future iPhone 18 silicon will reportedly triple on-device Transformer throughput, further reducing cloud fallback.

Gartner projects that by 2028, 40% of consumer AI queries will be processed in privacy-assured edge fabrics, such as PCC, up from 5% in 2024.

The upshot: With Private Cloud Compute, Apple reframes the AI conversation, your data stays yours, yet your assistant gets smarter. For enterprises and consumers alike, disruption now comes with end‑to‑end encryption.

––––––––––––––––––––––––––––

¹ Craig Federighi, WWDC 2025 Keynote, June 10 2025.