We have learned that ByteDance is developing an AI smartphone, currently named the "Doubao Phone," with ZTE acting as the ODM (Original Design Manufacturer) for production. The phone is expected to launch by the end of this year or early next year, initially for testing by ByteDance's internal teams, with no immediate plans for public sales.

According to our information, the main team behind the Doubao Phone is ByteDance's Ocean team, which handles AI hardware and reports to Flow, ByteDance's AI product division. Within Ocean, ByteDance is exploring multiple AI devices, including the Ola Friend smart earphones launched last year, AI glasses currently in development, and other undisclosed AI devices.

Ocean is led by Kayden, who reports to Alex Zhu, the head of Flow. Ocean's members include various hardware product teams that ByteDance has acquired over the years, such as Smartisan phones, PICO VR headsets, and Ola Dance smart earphones. Last year, the department also recruited mobile industry professionals who joined ByteDance after leaving Honor following Zhao Ming's departure as Honor's CEO.

ByteDance's phone-related venture began in the first half of 2024, around the same time the company acquired earphone manufacturer Oladance. Initially, ByteDance positioned itself as a large language model and AI supplier, seeking partnerships with phone manufacturers to integrate Doubao and other AI technologies as system-level capabilities. One potential partner was Nubia, a smartphone brand under ZTE.

However, this approach faced significant obstacles. A mobile industry analyst told us that almost all phone manufacturers ByteDance approached rejected collaboration. They view "large language models" as the "soul" of next-generation smartphones and want to develop in-house capabilities or maintain absolute control, unwilling to cede this to a powerful third party.

When selecting external large language model partners, phone manufacturers prefer smaller AI startups: OPPO has partnered with StepFun, while Zhipu has secured clients including Honor and Samsung. Every phone manufacturer is reluctant to surrender their autonomy in the large language model era.

An executive from an edge-side large language model startup told us that some phone manufacturers now compare the changes brought by large language models to how iPhone and Android phones once disrupted Nokia.

After encountering difficulties with major brand partnerships, ByteDance decided to develop its own phone with deep integration of the Doubao large language model. ByteDance handles the large language model functionality and some operating system-related work, while ZTE manages hardware design and production as the ODM.

ByteDance's integration of large language models into smartphones represents one of their comprehensive AI hardware initiatives. At ByteDance's company-wide meeting in February, CEO Liang Rubo outlined three major AI goals for 2025: exploring AI capabilities' upper limits, exploring new UI interaction formats, and strengthening scale effects. The "new UI interaction formats" include hardware platforms like smartphones.

ByteDance has attempted hardware products multiple times. In 2020, the education hardware team led by Yang Luyu launched products like the Dali Smart Lamp and writing tablets. The Dali Smart Lamp had a sales target of 2 million units in 2021, but sales fell far short of expectations. In 2021, ByteDance acquired VR headset brand PICO for 9 billion yuan, with PICO's team reaching over 2,000 employees at its peak. ByteDance invested heavily in VR content ecosystem development, spending 1 billion yuan on Qatar World Cup broadcasting rights for VR transmission and producing VR concerts featuring stars like Zheng Jun and Wang Feng.

In recent years, both education hardware and PICO teams have significantly downsized. However, major tech companies with AI ambitions cannot abandon hardware business, as hardware is considered a crucial direction for large language model implementation, especially smartphones.

ByteDance's entry into the smartphone business is most significant because it enables a "trinity" of terminal AI deployment—combining large language model capabilities, super apps, and hardware terminals. Currently, only Alphabet has assembled similar assets, controlling the Gemini series of large language models, Google Mobile Services essential for Android phones overseas, super applications like YouTube, and Pixel smartphones.

Simultaneously developing models, applications, and hardware allows for deeper integration, such as assigning physical buttons for AI functions, implementing more proactive AI interactions at the system level, or automatically providing AI voice summaries during phone calls, making services seek users rather than users seeking applications.

Tomorrow (August 20), Alphabet will officially unveil four smartphones: Pixel 10, 10 Pro, 10 Pro XL, and 10 Pro Fold. Media reports indicate the Pixel 10 series will feature multiple AI capabilities, including instant photo suggestions, voice image editing, and the "Pixel Sense" AI assistant. Pixel currently holds less than 1% of global market share.

An hour ago, ByteDance issued a response stating that Doubao is exploring opening AI capabilities to various hardware manufacturers, but all collaborations do not involve developing or launching proprietary smartphone products.