Apple restricts on-device AI to select hardware, subscription

At its WWDC 2026 keynote, Apple announced a rebuilt, Gemini-powered Siri AI and a next generation of Apple Intelligence, with Siri AI listed to arrive in English later this year (Apple). Multiple outlets describe it as Apple's most significant Siri overhaul in years, spanning iOS 27, iPadOS 27, and macOS "Golden Gate" (The Verge; TechRadar).

9to5Mac, MacStories, and TechRadar report that Apple's most powerful on-device model, and features it enables such as expressive voices and advanced dictation, is restricted to newer, higher-memory hardware: iPhone 17 Pro, iPhone 17 Pro Max, and iPhone Air; iPads with M4 or later; Macs with M3 or later; and Apple Vision Pro, each with at least 12 GB of RAM (9to5Mac; MacStories; TechRadar). Apple frames the memory floor as necessary to preserve latency, energy, and reliability for local model execution (MacStories).

Apple runs foundation models both on-device and through Private Cloud Compute, its server-side privacy tier (MacStories; The Verge). At a WWDC media event, Apple software chief Craig Federighi said, "We believe privacy in AI is non-negotiable," and said Apple uses none of the models Google deploys to its own customers (The Verge; The Next Web). Apple's new foundation models were custom-built in collaboration with Google's Gemini, and heavier generative features, including photorealistic image generation in Image Playground, rely on cloud models (TechCrunch; The Next Web).

Because server-side features such as image generation run on powerful cloud models, they carry daily usage limits that users can raise through most iCloud+ subscription plans (TechCrunch). Coverage also notes regional staggering: Notebookcheck reports Siri AI is initially restricted on iPhone and iPad in the EU while remaining usable on Macs set to English (Notebookcheck).

Apple is a major platform owner, so its hardware gates and delivery model shape developer expectations and purchase decisions. Many Apple Intelligence features still run on older hardware via cloud or Private Cloud Compute, while the highest-tier on-device capabilities are reserved for newer, higher-RAM machines, affecting latency, offline use, and feature availability. Apple's reliance on Gemini for parts of the stack also signals tolerance for third-party cloud dependencies rather than a fully in-house, on-device pipeline for every feature.

• •How Apple documents exact hardware and memory checks, and whether developers get APIs to detect on-device versus cloud-backed capability.
• •How iCloud+ usage caps affect teams building features that lean on frequent model-backed generation.
• •Regional rollout timing and any expansion of on-device support to additional devices.