The Great AI Wall: Apple is Building a Custom Siri for China
Toggle Dark Mode
While Apple Intelligence launched to a somewhat staggered rollout two years ago, one of Apple’s largest iPhone markets remains conspicuously absent from the party: China.
Even as Apple’s new AI features slowly expanded from a US-only deployment in iOS 18.1 to other English languages in iOS 18.2 and then even made it to the European Union in iOS 18.4, Apple struggled to launch Apple Intelligence in China due to unique regulatory requirements.
The crux of the problem is that while Apple’s US-made AI models were fine everywhere else, Beijing insisted that large language models (LLMs) on Chinese iPhones must be free of foreign influence. That means that Apple had to find an AI partner in China to co-develop its LLMs. Extensions like ChatGPT, which Apple relied on to back up Siri in iOS 18 (and iOS 26) would also be a no-go.
While the new Siri AI makes that second aspect far less crucial, it also gives Apple an even higher mountain to climb. Even in the rest of the world, Apple had to partner with Google for help in building its own foundation models, using the search giant’s work on Gemini as the scaffolding.
To be clear, the AI models are still 100 percent Apple’s, but Apple engineers leaned heavily on Google’s expertise to create them, and the nature of the new technology means Apple’s Private Cloud Compute servers may pivot to Nvidia chips rather than using Apple silicon, as originally envisioned.
While Apple finally found a partner in Alibaba and reportedly got the green light to proceed in China last month, it sounds like it’s going to be relying on a similar approach to build Siri AI for China — but from an entirely different starting point.
AI in The Middle Kingdom
Contrary to what some had thought, Apple won’t be using an off-the-shelf LLM from Alibaba to launch in China. Instead, sources speaking with Reuters report that Apple has trained an entirely new LLM specifically for the country.
That probably shouldn’t be too surprising, considering what we’ve already seen with the Google partnership, but it does fly in the face of previous reporting, that had implied that Apple would simply use a Chinese partner’s model through a method similar to the ChatGPT extension that’s been available in the rest of the world since iOS 18.2.
Although there was some reason to believe that Apple might be willing to consider such a compromise to get Apple Intelligence live in China sooner, that’s almost certainly not its preferred approach. Apple prefers to control the user experience, so developing its own LLM was always the ultimate goal, and the work it’s done in getting Siri AI ready for iOS 27 has undoubtedly also empowered it to follow suit by developing and training a similar model with Alibaba.
Still, the other side of that question was always whether such an arrangement would be palatable to Beijing. Every AI model available in China right now is fully owned and developed by Chinese companies, most of which are influenced, if not outright controlled, by the state. As Reuters notes, Apple could end up being the first foreign company to receive approval for a proprietary AI model in China.
Previous reports have also suggested Alibaba’s Qwen model will be incorporated into Apple Intelligence. While it remains unclear exactly how that will work, it’s likely to be the China-equivalent of the ChatGPT extension. Apple may also be using the Qwen frameworks to build its own LLMs, much like it leveraged Gemini in the rest of the world.
There’s still no word on when Apple Intelligence will actually launch in China. The government approval clears the path, so it could arrive as soon as iOS 27.0 next month, but the complexities of getting all the pieces in place could see it pushed to a later point release, much the same way the original Apple Intelligence features didn’t begin rolling out until iOS 18.1.
