Apple's late entry to AI

While OpenAI launched ChatGPT in 2022 and Google reacted with Bard, Apple was conspicuously absent from the public AI race until WWDC 2024, when it announced Apple Intelligence. The Apple bet wasn't the most capable model — it was the most private one.

The on-device approach

Apple's on-device models are small (~3B parameters) but optimized to run on the Neural Engine of iPhone 15 Pro and later. They handle: writing tools (rewrite, summarize, grammar), smart reply in Mail and Messages, basic image generation, Siri queries that don't need broad knowledge.

Private Cloud Compute: the architecture

For tasks that exceed on-device capability, Apple invented Private Cloud Compute (PCC): servers in Apple data centers but with a security architecture that cryptographically guarantees Apple cannot access the data being processed.

The chain: device encrypts data with ephemeral keys, sends to PCC node, node processes, returns response, the data isn't logged anywhere. Apple opened the architecture for external auditing — uncommon move in cloud services.

The ChatGPT integration

For tasks that exceed Apple Intelligence (broad questions, complex creativity), Apple integrated ChatGPT via opt-in. The user has to explicitly accept sending the request to OpenAI. It's the most pragmatic option to cover what its own models can't do.

In iOS 27 (May 2026), Apple expands this concept to Extensions, opening Claude and Gemini too. The shift: Apple no longer pretends to do everything in-house.

Limitations and reality

The market response to Apple Intelligence was lukewarm. Reasons: (1) on-device models are limited (3B can't compete with GPT-4o), (2) the renewed Siri kept being delayed, (3) compelling consumer use cases never appeared. Apple is rectifying with iOS 27.

Conclusion

Apple Intelligence isn't the best AI — but is the most private and the one with most strategic distribution (every iPhone). For users with high privacy concerns, it's the default option. Apple's 2026 challenge: bridge the gap with frontier labs without giving up privacy posture.