Apple split its AI features between the device and its own cloud

An architecture claim, announced before a shipped feature

Apple's 10 June 2024 announcement introduces Apple Intelligence as a set of writing, image and assistant features across Mail, Notes, Messages and Siri, but the substance of the announcement is architectural rather than feature-by-feature: Apple states that “many of the models that power it run entirely on device,” with more demanding requests handed to a system called Private Cloud Compute, described as running on Apple-designed servers with independent code inspection offered as a check on the claim. Systemwide Writing Tools for rewriting, proofreading and summarising, notification summaries, and an Image Playground generator are named as the first features built on this split.

Who could actually use it, and when

The announcement restricts the feature to iPhone 15 Pro or Pro Max and to iPad or Mac models with an M1 chip or later, running the then-new iOS 18, iPadOS 18 or macOS Sequoia, with Siri and the device language set to US English, a narrower list than “the next iPhone update” might suggest. Availability itself was staged: a beta “this fall,” in US English only, with other languages and platforms following “over the following year,” and a separate, opt-in integration with ChatGPT through Siri arriving “later this year.” A reader in mid-2024 comparing the announcement to their own device could conclude the feature existed and still not have access to it for months.

What the split is meant to solve, and what it does not resolve

The stated reason for the on-device and Private Cloud Compute split is privacy: keep what can be kept local, and make what has to leave the device independently checkable rather than opaque. Apple's own later Apple Intelligence product page extends this framing considerably further, describing a much larger current feature set including a conversational Siri and camera-based Visual Intelligence, with tighter minimum-hardware requirements for some capabilities than the 2024 launch required. None of Apple's own material addresses what happens when a request is ambiguous about whether it needs cloud processing, or what a user sees differently, if anything, when their request has left the device.

Questions to ask before you adopt it

• Does your specific device meet the hardware and OS requirements, or only a newer device in the same product line?
• For a given feature, do you know whether it runs on-device or is sent to Private Cloud Compute?
• If a Siri request is routed to a third-party model such as ChatGPT, is that routing something you opted into?

Apple's announcement is unusual in this batch for leading with architecture rather than a feature list, and that architecture is precisely what a reader has the least way to verify from the outside. The independent inspection Apple offers for Private Cloud Compute is the mechanism meant to close that gap; whether it does is a separate, ongoing question this record does not settle.