Apple Intelligence launched in 2024. It still isn't running on a single iPhone in mainland China, and until Wednesday, nobody outside the two companies knew exactly when that would change. On July 15, 2026, China's Cyberspace Administration registered Apple's generative AI service, clearing the specific regulatory hurdle that had kept Apple's flagship AI features locked out of its second-largest market for nearly two years.

Alibaba confirmed the arrangement directly to CNBC: its Qwen model will power Apple Intelligence across iOS, iPadOS, macOS, and visionOS for users in China. No launch date came with the announcement. Investors didn't wait for one. Alibaba's US-listed shares jumped as much as 7.5% during the session, after opening roughly 4% higher in premarket trading, while Apple shares climbed to a new all-time high the same day.

A Regulatory Filing Two Years in the Making

China requires any company offering large language models or generative AI services to the public to register with the Cyberspace Administration before launch. That's not a formality Apple could route around, and it isn't a rule unique to foreign companies either. It's the same registration every domestic AI provider in China has to clear. What made Apple's case take this long wasn't the paperwork. It was finding a model partner whose technology could satisfy both Apple's product standards and Beijing's compliance requirements at the same time.

According to Reuters, the approved version of Apple Intelligence will incorporate capabilities from AI models built by both Alibaba and Baidu, not Alibaba alone. Alibaba's Qwen appears to be the more prominent partner based on its own public statement, describing the integration as giving Chinese users access to text and image understanding and generation without leaving Apple's interface. Baidu's specific role in the final product hasn't been detailed publicly yet.

The Partners Apple Tried and Passed Over

This deal didn't happen on the first attempt. Before settling with Alibaba, Apple reportedly explored partnerships with Baidu on its own, along with DeepSeek and models from ByteDance, according to TechCrunch. Those earlier conversations reportedly ran into difficulty adapting the models to the specific requirements of Chinese customers, and that friction is a big part of why a product that launched everywhere else in 2024 is only now clearing its final hurdle for China two years later.

That timeline says something about how differently the AI partnership calculus works inside China compared to Apple's approach elsewhere. In the US and most other markets, Apple built its own foundation models with Google's infrastructure behind them, an arrangement detailed in Apple's AFM 3 announcement earlier this year. In China, foreign model access is restricted enough that Apple had no equivalent option. It had to find a homegrown model good enough to meet its bar and compliant enough to satisfy regulators, and that search apparently took multiple failed pairings before Qwen worked.

What the Deal Actually Covers

The integration, as Alibaba described it, sits at the interface level rather than replacing Apple's own software. Users get access to Qwen's capabilities, specifically text and image understanding and generation, from directly within Apple's system experience rather than through a separate app. That distinction matters for how Apple maintains its product identity. The company isn't handing Chinese users a rebadged Qwen chatbot. It's routing specific AI capabilities through Alibaba's model while keeping Apple's own interface, design language, and system integration in place, the same architecture pattern Apple used with Google elsewhere.

For Alibaba, landing inside hundreds of millions of iPhones is a different kind of win than topping a benchmark leaderboard. Model distribution at that scale, embedded in the operating system rather than requiring a separate download, is exactly the kind of real-world deployment that turns technical capability into actual usage. Qwen becomes the default AI experience for a huge population of iPhone owners who will likely never know, or care, which company built the model running underneath Apple's branding.

Why Apple Couldn't Afford to Keep Waiting

The financial stakes behind this delay were substantial. Apple's sales in Greater China rose 28% to $20.5 billion in its most recent quarter, and the company recently reclaimed the number two position in China's smartphone market after a shopping festival built around iPhone discounts. Every quarter Apple Intelligence stayed unavailable in China was a quarter Apple's flagship AI feature simply didn't exist for one of its most important customer bases, at exactly the moment domestic Chinese phone makers have been shipping their own AI features aggressively.

A Smaller Story Tucked Inside the Bigger One

One additional detail is worth watching independent of the regulatory news itself. CNBC reported separately that Apple is in talks with PrismML, a Caltech spinout backed by Khosla Ventures, whose engineers have already released a compressed version of Alibaba's open-source Qwen model capable of running entirely on an iPhone 15 or newer, all 27 billion parameters intact. That's a meaningfully different technical achievement from the cloud-routed integration Alibaba just confirmed. If Apple eventually adopts on-device compression techniques like PrismML's, the question stops being just "which company's model powers Apple Intelligence in China" and becomes "how much of that model can run locally, without a network connection at all." That's a much bigger shift, and this week's regulatory approval is likely just the first visible step toward it.