When Will Apple’s AI Transformation Catch Up with Global Standards?

Written by | Zhao Weiwei

Early excitement once boosted Apple’s stock price, but with no clear launch date for the official version of Apple Intelligence—delayed for two years—the company is expected to release a ‘beta’ version this September, initially excluding China and the EU. This falls far short of user expectations.

The 2026 Apple Developer Conference ended with Apple’s stock price dropping nearly 2%, marking its largest single-day decline in nearly three weeks—a direct consequence of these issues.

“As long as privacy is prioritized, I don’t mind Siri evolving more slowly. If I want speed, I’ll switch to Android and hand all my data over to Google,” read a popular comment beneath a related Wall Street Journal article.

But users quickly retorted: Apple has already partnered with Google, and the next-gen Siri will rely on Google’s Gemini model and cloud infrastructure. High-level Siri requests will be routed to Google’s servers, meaning Apple’s privacy safeguards cannot sidestep Google entirely.

The crux of the debate: “Google pays Apple $20 billion annually just for default search engine placement. This reveals how valuable user data is to Google—if Apple truly protected privacy, why would Google keep paying?”

Such sentiments reflect widespread disappointment among Apple users, who question not only technical shortcomings but also a growing trust deficit between the two companies.

Though Apple emphasized at the event that “AI privacy is non-negotiable,” the new Siri is a collaborative effort with Google, with Apple licensing the Gemini model and building its own framework atop it.

The 2026 Apple Developer Conference also marked Tim Cook’s final act as CEO. Under his leadership, Apple became the world’s most profitable consumer electronics giant, with services creating an unassailable competitive moat.

Yet this empire has been sluggish in responding to today’s AI wave, forced to rely on a rival’s model to construct its own AI system.

John Ternus, promoted from the Mac and iPhone development teams, will succeed Cook as CEO. Facing Google, Microsoft, Meta, and others’ fortified positions in software and AI, Ternus must address the question Cook left unresolved: When will Apple’s AI transformation align with global industry standards?

1. Laying the Groundwork for a Delayed Siri

Siri AI is Apple’s revamped voice assistant, powered by the new Apple Intelligence platform—a joint venture with Google.

It features a refreshed design and voice, with a Siri bubble appearing at the top of the screen. Users can customize Siri’s expression and speech speed. Siri now has screen awareness, can interpret on-screen content contextually, and includes built-in image editing and writing tools. Related functions sync across devices, with interaction records stored in the new Siri AI app.

The latest iOS 27 on mobile devices lets users activate Siri by swiping down from the Dynamic Island; the camera app introduces a “Siri Mode” for real-time object identification and information display.

On Macs, the biggest change is Siri’s direct integration into the Spotlight search bar.

Users can ask questions directly in the search bar or query specific files, eliminating the need for a separate assistant interface. Like iOS 27, macOS 27 introduces liquid glass transparency sliders, refined window corner radii, and restored colored sidebar icons.

Notably, upgrades to Apple’s built-in browser further integrate AI capabilities.

For example, the new AI tab management feature automatically organizes open web pages by theme. The “Notify Me” function lets Safari monitor webpages for changes, such as alerting users when a restocked product becomes available. Additionally, users can describe their needs in natural language, and Safari will generate browser extensions accordingly.

For Apple, AI’s deeper value lies in cross-app integration.

For instance, the Messages app will proactively suggest shareable photos based on conversation context; the call interface will display relevant details like flight confirmation codes; the Mail app adds intelligent content suggestions and supports natural language scheduling synced to the calendar. System shortcuts also allow natural language creation, while the Image Playground offers composition adjustments and enhanced “distraction removal” for AI-level photo editing.

Of course, Apple caters to all users, not just AI enthusiasts. For younger users, Apple has optimized parental controls—a key concern for parents.

The child account system has been overhauled, allowing parents to precisely select which apps and content their children can access, requiring authorization before opening new websites. The Screen Time menu interface has also been redesigned for clearer navigation, enabling parents to manage device usage more efficiently.

It’s evident that Apple’s system-wide upgrades are paving the way for Siri AI’s debut.

2. iPhone 16 Users Feel Let Down

This marked Cook’s final appearance as CEO at Apple’s WWDC.

Instead of summarizing his tenure with data, he stated: “Creating the world’s best products and enriching lives has always been our mission. It has been my lifelong honor to work with such creative partners and integrate technology into daily life.”

In contrast, the event was dubbed “the most disappointing WWDC ever.”

“I’ve followed Apple’s events since the early 2010s, and this is by far the worst,” read a post on the Apple enthusiast forum MacRumors. “So they’re really doing another non-live demo? Isn’t this what messed them up in 2024?”

At the 2024 WWDC, the Siri functions showcased were not running in real-time on actual devices but were pre-recorded videos. Even the Siri team hadn’t seen a fully functional official version at the time. Apple later faced a class-action lawsuit for false advertising and settled out of court for $250 million in 2025.

Two years later, where do users stand on the new Siri?

Comments on Reddit reflect the frustrations of many Apple users. One highly upvoted comment read: “I’m really disappointed. My iPhone 16 Pro Max can’t run the most powerful on-device Foundation Model.”

The iPhone 16 Pro Max, Apple’s flagship model released in September 2024, was marketed as AI-centric, claiming the iPhone 16 was the first model built specifically for Apple Intelligence.

A year and a half later, Apple’s AI features remain partially implemented. While this model can use Siri AI online, it cannot leverage Apple’s most powerful on-device model.

An on-device model runs the AI directly on the phone’s chip without an internet connection, keeping all data local—a key privacy advantage Apple promotes. Running Apple’s most advanced on-device model requires 12GB of memory, a threshold currently met only by the iPhone Air (12GB) and iPhone 17 Pro/17 Pro Max (12GB).

In other words, users who bought the iPhone 16 and iPhone 17 standard models for their AI capabilities now feel betrayed by Apple, excluded from the top-tier AI experience.

This is the dilemma Apple must confront: Its AI promises are eroding user trust. How will the brand fulfill past commitments to consumers? How can Apple’s product replacement cycles keep pace with the rapid iteration of AI technology and its promises to users?

Apple markets current hardware with next-gen AI aspirations, but reality catches up, and hardware thresholds rise to the next generation.

3. Cook’s Unanswered Questions

“If I were Ternus, I’d let Cook swallow all this and make my grand entrance next year.”

This sentiment, prevalent on U.S. tech forums, leans toward conspiracy but also indirectly confirms Ternus’s absence from this WWDC stage. Rumors suggest Apple’s new CEO will host the iPhone 18 event in September, which will be his true debut.

For Cook, the AI battle has been a compromise.

Foreign media previously reported that Apple signed a multi-year licensing deal with Google, using a customized Google Gemini model (about 1.2 trillion parameters) to power the cloud-based advanced functions of the new Siri, with annual costs around $1 billion. Apple also incurs additional computing expenses.

For comparison: In May 2026, SpaceX’s IPO filing revealed that Anthropic and xAI had partnered, spending $1.25 billion monthly to rent over 220,000 NVIDIA GPUs at the Colossus data center, totaling $15 billion annually.

Thus, in the AI race, Cook opted for a cost-priority strategy: avoiding the large model arms race, Apple spends $1 billion annually to lease Google’s model instead of investing $15 billion like Anthropic to develop its own. At this stage, building cutting-edge large models costs far more than leasing, with a significant gap.

Apple and Google have now jointly developed five models, categorized as on-device and cloud-based. One on-device model, AFM 3 Core, is an upgraded 3 billion-parameter general model with significantly improved capabilities.

The AFM 3 Core Advanced is currently Apple’s most powerful on-device model, natively supporting text, images, and voice. It offers more natural voice interactions and higher speech recognition accuracy. With 20 billion parameters, it adopts a sparse architecture, activating 1 to 4 billion parameters based on demand, delivering outstanding efficiency and deep optimization for Apple’s flagship chips.

On the private cloud, Apple also operates three models, with promises that user data will not be retained or shared with anyone, including Apple:

1. AFM 3 Cloud: The primary cloud model, balancing speed, efficiency, and performance;

2. ADM 3 Cloud (Image Model): Focused on image generation and editing, enabling advanced retouching and creative image creation;

3. AFM 3 Cloud Pro: The most powerful cloud model, designed for complex tasks like intelligent operations and deep logical reasoning.

Among these, AFM 3 Cloud Pro is deployed on Google Cloud’s NVIDIA GPU clusters, making it the model with the highest degree of Google involvement.

When Apple users invoke AFM 3 Cloud Pro for Siri functions, Apple’s privacy system remains active: Data is anonymized before transmission, and the agreement explicitly prohibits Google from using the data for model training, though the computation relies on Google’s infrastructure.

It’s as if Apple built an AI skyscraper, with Apple security guards, Apple cameras, and Apple locks. But the steel came from Google, and the most luxurious penthouse was built on Google’s foundation.

This structure is the AI legacy Cook left for Ternus—a reflection of Apple’s late start and slow pace in AI infrastructure investment, but also a phased achievement. Apple now has a defensive position in the AI battle, even if it was built jointly with a rival.

The problems Cook didn’t solve during his tenure now sit untouched on Ternus’s desk. The AI-era narrative has just begun, and while it must keep pace with the world, its author has changed.

Reviewed by | Chen Qiulin

END