Following the National Day holidays, which were filled with discussions about new flagship devices from Huawei, a more interesting debate in the country's phone market centers on who will control the layer of artificial intelligence situated between users and their applications. Commentary from the technology section of Internet Jianghu, published in Tencent News on October 9th, summarizes the positions of the four largest players: Huawei integrates deeply, Alibaba operates broadly, ByteDance focuses, and Tencent acts subtly.
Huawei is the only company that creates its entire technology stack independently. Its proprietary chips, operating system, and models allow it to implement AI at the lowest levels of the phone. The article notes a more powerful NPU in the Kirin 9050 Pro, capable of running a 30-billion parameter multi-model model on the device. Huawei relies on its own models and connects to the WeChat assistant only via the A2A agent protocol. The advantage is higher potential, including offline capability and lower power consumption. However, this involves a very long supply chain that few competitors can replicate.
ByteDance chose a narrower path with the Doubao phone, built on Nubia from ZTE using the second-generation NaviX Ultra. The new version invokes applications through MCP and A2A protocols instead of screen tap simulations, but fully automated actions are mainly achieved within ByteDance's own services.
Alibaba offers a platform. Its Qwen Intelligence solution provides phone manufacturers with model capabilities, while they handle the system level and real users, as Honor did with the Magic9 and Robot Phone series. If the model works there, it can be replicated on other Android brands at relatively low cost.
Tencent does not compete for the system agent at all. Through A2A partnerships with Huawei, Xiaomi, Honor, OPPO, and vivo, phone assistants send instructions to WeChat, which executes them. Separately, the company collaborates with Qualcomm to run 7B and 3B Hunyuan models on Snapdragon 8 Elite devices.
The commentary asserts that the real barrier is the ecosystem. Phone manufacturers fear becoming mere hardware channels. Model developers earn revenue from APIs but do not gain user relationships. Super-apps like WeChat, Taobao, and Meituan are simultaneously agents that services must call upon and potential competitors, and the first Doubao phone faced limitations from major applications. The author concludes that the industry needs a new revenue distribution method based on completed user intent, rather than advertising traffic or token count, before any company can claim the first point of AI entry in the phone.
