The Apple China page at apple.com.cn/apple-intelligence/ states that the first batch of Apple Intelligence core features ships "later this year" on fifteen iPhone models, from iPhone 15 Pro (A17 Pro) up to iPhone Duo (A20 Pro). The page names Writing Tools, Genmoji, Visual Intelligence, real-time translation across Messages, FaceTime, and Phone, and a Siri that accepts typed input from anywhere on the system. It also lists a "Remove" tool for photos, a custom memory-film generator that assembles clips from a text prompt, and a priority-notification mode that stacks top alerts as summaries above the rest.

It says "今年稍晚" and stops. This is a country homepage, not a release note, and the absence of a date here is not a contradiction. It is the vendor's floor statement. The wire's specificity is an addition layered on top, not a correction of what the page says.

The constraint is the silicon floor. A17 Pro is the oldest chip on the list, which means every iPhone 15 non-Pro, every iPhone 14, and every iPhone 13 is excluded from Apple Intelligence in China. Apple is not shipping this feature set to A16 or A15 silicon, and the page does not explain why the floor sits where it does. The Qwen integration for Siri, Visual Intelligence, and Writing Tools is the China-specific delta that the US page does not carry. For the buyer in mainland China, the feature set is not identical to the US build, and the page does not say which operations stay on-device and which route through Qwen. That routing question is the one the spec table cannot answer, and it matters because the privacy claim on the page is absolute: no one, including Apple, accesses user data. If Qwen is in the loop, that claim needs a boundary the page does not draw.

No benchmark, no latency figure, no supported-language list, no battery-impact number. The page claims device-side processing and says no one, including Apple, can access user data, but it does not state which inference steps run on the Neural Engine versus which pass to Qwen. The "later this year" window is the only temporal anchor, and it is not a date. The fifteen-model list is a compatibility floor, not a launch schedule, and the page does not say whether all fifteen models receive the same feature set on the same day.

The page lists iPhone Duo, iPhone 18 Pro Max, and iPhone 18 Pro as the top of the stack, all on A20 Pro silicon. Those are not yet shipping in any market. Their presence on the compatibility list means Apple is pre-clearing the next generation for Apple Intelligence, but the page does not say whether those models launch with the feature set active or whether they are placeholders for a future update.

The A17 Pro floor excludes every non-Pro iPhone 15 and all older generations from Apple Intelligence in China, and the Qwen routing for Siri and Visual Intelligence creates a privacy boundary the page states but does not define. The buyer cannot tell from this page which inference steps stay on the Neural Engine and which leave the device.

The silicon floor at A17 Pro means Apple Intelligence in China requires the 3 nm N3B process minimum. The Qwen integration for Siri, Visual Intelligence, and Writing Tools is a China-only routing path absent from the US build, and the page's absolute privacy claim (no one accesses user data, including Apple) is in tension with that external model call. The page does not partition on-device versus Qwen-routed inference, so the privacy guarantee is unverifiable from the document.