Apple has stumbled in its rollout of Apple Intelligence and is now in talks with OpenAI and Anthropic about powering Siri with either ChatGPT or Claude, to potentially be launched next year. Apple’s partnership with Alibaba in China, which was announced in February, might give us hints about how Apple will approach outsourcing its large language model technology globally.
At the Worldwide Developers Conference in June 2024, Apple announced that AI-enabled Siri would soon be able to provide more fully contextualized responses based on the personal data and on-screen activities of iPhone users. This rollout was originally planned for Fall 2024 but was delayed once and then delayed again until sometime in the “coming year.” Apple has faced technical difficulties rebuilding Siri from scratch to make it fully AI-capable and reports detail a series of management shakeups as the company figures out how to keep pace with AI.
But the narrative that Apple has failed to develop more robust in-house AI capabilities might only be part of the story. Given that Apple is already partnering with Alibaba to bring Apple Intelligence to China, Apple’s decision to outsource its foundational AI model for Siri to OpenAI or Anthropic seems strategically consistent with this prior choice.
Apple Intelligence was released in select countries last year to enhance Siri, provide writing assistance, and generate emojis, although this is only part of what Apple promised in terms of AI-enabled functionality. In China, Apple Intelligence is not yet available in any form due to regulatory constraints.
Before any AI in China can be deployed it needs to be vetted first by the Cyberspace Administration of China and receive final approval from the State Council. China is particularly concerned about storing data and censoring content, including query responses created by Siri. Blocking access to politically sensitive content helps the Chinese Communist Party stifle collective action and maintain regime stability.
Apple’s decision to outsource its large language model technology to Alibaba is the only reason why it will be possible for the company to provide Apple Intelligence to customers in China. According to the Financial Times, a top Beijing official warned foreign companies that it would be a “difficult and long process” to release AI in China unless they partner with local AI companies.
While Apple has not yet officially confirmed the partnership with Alibaba, the deal was first reported by The Information in February and confirmed by Alibaba’s chairperson, Joseph Tsai (蔡崇信), shortly after. Alibaba announced in June that Qwen3, Alibaba’s latest LLM, has been optimized for Apple’s AI architecture, signaling that the rollout might be imminent. But reporting from last month also indicates that trade war tensions between the U.S. and China are stalling approval by the Cyberspace Administration.
Time is of the essence. Apple’s sales in the Greater China region have declined for two straight years. Apple faces stiff competition from local competitors including Huawei, which despite being hindered by U.S. export controls, has rebounded impressively. While Apple and Huawei respectively claimed 70% and 13% of the high-end smartphone market in China in 2023, Apple now has only 47% market share and Huawei has jumped to 35%. Last year, Huawei launched the Mate 70 smartphone series, which integrates in-house AI features.
Outside of China, Apple is similarly feeling an urgent pressure to upgrade its AI capabilities but seems to lack the internal capacity to do so in a timely manner, thereby making it strategically consistent for Apple to outsource its LLM technology the way it has in China. Axios called Siri a “glaring anachronism” because its functionality as a voice assistant pales in comparison to the best chatbots on the market. Market perception that Apple is behind in the AI race is tanking its stock, which has dropped 15% this year. While the iPhone certainly still reigns supreme, Apple’s global competitors, like Samsung and Google, are integrating AI features into their smartphones at a quicker pace. In June, Bank of America analysts estimated that Apple was two or three years away from deploying a generative AI powered Siri. Bloomberg’s Mark Gurman gave a similar estimate.
There are also some similarities in how Apple has approached AI partnerships both in and outside of China. In both contexts, Apple has some flexibility in picking which large language model best suits their goals and privacy standards. Apple’s AI system includes both on-device models deployed locally and Private Cloud Compute that allows users to do more complex AI-powered tasks on models hosted on the cloud. As stated on Apple’s website, these AI features are deployed “all while protecting your privacy.”
In China, Apple originally partnered with Baidu to integrate Ernie 4.0 into Siri, but ran into several problems including that Baidu’s models struggled to understand prompts and provide accurate responses, according to the Information. Baidu and Apple also clashed on privacy — Baidu wanted to use data collected from iPhone users to train its models but Apple’s privacy policy forbids this. Apple was also in talks with ByteDance and Tencent about using their technology to power Apple Intelligence in China.
In the American context, Apple certainly has less flexibility than it would like. Tech expert Ben Thompson argues that Apple’s flailing AI efforts have hurt its leverage in outsourcing a model. Since Apple needs help with foundational AI capabilities, they need to “pay up.” Still, Apple is taking the time to figure out which model is best to power Siri. According to Bloomberg, Apple asked both OpenAI and Anthropic to train versions of their models suitable for Apple’s Private Cloud Compute infrastructure. This allows Apple to evaluate which model not only provides the best user experience, but also integrates seamlessly into Apple’s privacy-first AI architecture.
Apple obviously did not intend to fail at developing its own AI. But its response to regulatory headwinds in China and technical headwinds globally still lends a sense of strategic consistency.
Leave a Reply