Apple Intelligence coverage focused on Siri, on-device models, private cloud compute, AI features across Apple devices, and the privacy tradeoffs shaping consumer AI.
Stories
414
Latest source update
July 15, 2026
Coverage
Live
Topic brief
What to know about Apple Intelligence
Brief updated Jul 10, 2026
Apple Intelligence is Apple's brand for the AI features built into iPhone, iPad, Mac, Watch, and its wearables. It spans the Siri assistant, on-device and server-based Apple Foundation Models (AFM), generative tools in Photos and other apps, and the Private Cloud Compute system that runs larger models while aiming to preserve privacy. Apple's strategy leans on a hybrid of small on-device models running on Apple silicon and larger models in a hardened cloud, positioning privacy and tight hardware-software integration as its differentiators rather than raw model scale.
For practitioners, Apple Intelligence matters because Apple sets defaults for well over a billion devices. Mobile and app developers care about the Siri and App Intents surfaces, on-device model APIs, and how much third-party access Apple actually grants. ML engineers watch Apple's on-device model compression, foundation-model releases, and silicon roadmap, because efficient edge inference is a hard problem Apple is forced to solve at scale. Hardware and supply-chain teams track Apple silicon, custom ASICs, and memory sourcing, since Apple's component decisions move markets. Privacy, security, and compliance teams study Private Cloud Compute and Apple's consent design as a reference model for on-device and confidential AI.
Apple's approach is also distinctive because it now blends in-house models with an outside partner. Apple has historically emphasized doing AI on-device for privacy, but it has entered a multi-year collaboration to base parts of its stack on Google's Gemini models and cloud, a notable shift for a company that markets independence. That creates tension between Apple's privacy branding and its reliance on partners and cloud inference, and it puts Apple in negotiations with regulators, especially in the EU under the Digital Markets Act, over how and where these features can ship.
What changed recently
In the weeks since WWDC 2026, Apple has moved its Siri AI overhaul from keynote to the messier work of shipping, and the details reveal both ambition and constraint. Apple confirmed that Siri AI is rebuilt on its third-generation Apple Foundation Models and, notably, on a multi-year Google Gemini partnership, and it rebuilt the search infrastructure behind Spotlight, Photos, and Mail with an on-device semantic-search API. But the rollout is gated: beta testers found Siri AI can pull live data from only a narrow set of third-party apps with per-app permission, Apple Intelligence for Home sits behind a costly iCloud+ tier, and Apple began showing consent popups when some prompts are routed to Google Cloud, exposing the friction between its privacy branding and its cloud dependencies. Apple is also reportedly eyeing PrismML compression to run large models such as a 27-billion-parameter Qwen variant on-device, a sign it still wants to push inference to the edge.
The hardware and supply-chain picture hardened at the same time. Apple accelerated its silicon roadmap toward an M7 family focused on on-device AI, extended a custom-ASIC deal with Broadcom through 2031, and continued to dominate edge-AI smartwatch shipments. Yet an AI-driven memory shortage is squeezing Apple directly: it raised prices on Macs and iPads, is testing banned CXMT DRAM for China-market devices to manage supply, and analysts tie its margins to DRAM costs. On policy, Tim Cook held a constructive call with the EU over the standoff blocking Siri AI from launching in the bloc, even as the feature is set to ship globally with iOS 27 and iPadOS 27 in September. The throughline is that Apple's AI is now real and imminent, but hemmed in by partners, components, and regulators.
What to watch
The near-term calendar centers on Siri AI shipping with iOS 27 and iPadOS 27 globally in September, though its EU availability hangs on an unresolved Digital Markets Act standoff that Tim Cook and EU officials are still negotiating. Watch how much third-party app access Apple actually opens beyond the narrow beta set, whether reported interest in PrismML compression turns into on-device large-model support, and how the Google Gemini dependency evolves as the next generation of Apple Foundation Models is built on it. On hardware, Apple's M7 silicon family, its custom-ASIC deal with Broadcom running through 2031, and its in-development AI AirPods, pendant, and N50 smart glasses are the key roadmap items, while memory-supply moves such as testing CXMT DRAM for China signal how Apple plans to manage the AI-driven component crunch that is already raising device prices.
Comparison
tier
notes
example
On-device models
Run locally on Apple silicon for privacy and low latency
AFM 3 Core and AFM 3 Core Advanced
Private Cloud Compute
Apple's privacy-focused server tier for larger requests
AFM 3 server models
Partner cloud
Used for some features; Apple shows a consent popup when prompts are sent
Google Gemini via Google Cloud
Frequently asked questions
What is the difference between Apple Intelligence, Siri AI, and Apple Foundation Models?+
Apple Intelligence is the umbrella brand for Apple's AI features across its devices. Siri AI is the rebuilt assistant, unveiled at WWDC 2026, with conversational responses, on-screen awareness, and visual intelligence. Apple Foundation Models (AFM) are the underlying models; the third generation includes on-device models (AFM 3 Core and Core Advanced) and server models that run on Private Cloud Compute. In short, AFM is the engine, Siri AI is the assistant, and Apple Intelligence is the brand.
Does Apple's Siri now use Google's Gemini?+
Yes, at least in part. Apple has entered a multi-year collaboration under which the next generation of Apple Foundation Models is based on Google's Gemini models and cloud technology, and the Gemini-powered Siri was confirmed at WWDC 2026. Apple still runs smaller models on-device and larger ones on its Private Cloud Compute, but the partnership marks a notable shift for a company that markets AI independence, and Apple shows a consent popup when some prompts are routed to Google Cloud.
When and where will Siri AI be available?+
Apple has said Siri AI ships with iOS 27 and iPadOS 27 globally in September. Availability in the European Union is uncertain because of a Digital Markets Act standoff; Tim Cook and EU officials held a constructive call in late June about the block, but it was not resolved. Device compatibility is limited to newer hardware, and some features, such as Apple Intelligence for Home, require a paid iCloud+ tier.
How much can third-party apps integrate with Siri AI?+
So far, less than the marketing implies. In the iOS 27 betas, testers found Siri AI could pull live data from only a narrow set of third-party apps, with early examples like Tessie, Ford, and EV battery status, and Siri asks permission before accessing app data. Apple has also hidden some third-party Siri Extensions in beta builds. The broad third-party ecosystem Apple has described is not yet fully open, so developers should plan for a phased rollout.
Why are Apple's device prices going up, and what does AI have to do with it?+
The AI build-out across the industry has driven a memory-chip shortage, and Apple has cited memory and component costs as it raised prices on select Macs and iPads. Apple is managing supply in part by testing CXMT DRAM for China-market devices and by adjusting its silicon roadmap. For buyers, the practical effect is that AI infrastructure demand elsewhere is showing up in the price of consumer Apple hardware.
What is Apple doing to keep AI on-device and private?+
Apple emphasizes a hybrid model: small models run on Apple silicon on the device, and larger requests go to Private Cloud Compute, which Apple designs to preserve privacy. It has rebuilt on-device search for Spotlight, Photos, and Mail with a local semantic-search API, is reportedly interested in PrismML compression to run large models such as a 27-billion-parameter Qwen variant on an iPhone, and shows consent prompts when data leaves for partner clouds. On-device inference and Private Cloud Compute are central to how Apple differentiates its AI.
