CITIC Securities: Volcano Engine is empowering multi-category hardware products with AI implementation, with a focus on ByteDance ecosystem companies

Zhitong
2025.06.13 00:45

CITIC Securities released a research report indicating that ByteDance's Volcano Engine showcased the development trends of multi-category AI hardware products at the Force2025 Power Conference, emphasizing the widespread application of AI in hardware. The report focuses on companies within the Byte ecosystem and related component manufacturers. It is expected that products such as AI headphones and AI toys will become market highlights, with AIoT product shipments likely to exceed 10 million units by the end of the year

According to the Zhitong Finance APP, CITIC Securities released a research report stating that on June 11, ByteDance's Volcano Engine held the Force2025 Original Power Conference. In addition to the main forum, Volcano Engine also hosted product-specific forums, industry-specific forums, and partner forums. Through this conference, it was observed that Volcano Engine is empowering the AI implementation of various hardware products, and multimodal visual understanding applications are a clear trend. Looking ahead, there is optimism about the widespread implementation of AI in hardware, with AI glasses and AI toys expected to become the most typical products. It is recommended to focus on companies within the Byte ecosystem and component and module companies that are growing alongside brand clients.

The main viewpoints of CITIC Securities are as follows:

Byte empowers AIoT with multiple categories, emphasizing multimodal implementation.

At the Volcano Engine Force2025 Original Power Conference venue, various smart hardware products were showcased, including AI alarm clocks, AI learning machines, AI toys, AI mice, AI glasses, AI watches, AI headphones, AI mattresses, AI pillows, AI security cameras, AI PCs, AI phones, AI pets, and robots, indicating that large models are extending to multiple categories in hardware. Mr. Xing Xiaoci, head of smart hardware at Volcano Engine, introduced the advantages of the Volcano large model using AI headphones and AI toys as examples, specifically highlighting that AI toys can easily configure roles, tones, and action controls, while AI headphones based on ASR + Doubao large model return an average of 1.5 seconds for the first word, both achieving quick turnkey delivery. According to Mr. Xing Xiaoci, as of June 11, the shipment volume of AIoT products connected to Doubao exceeded 1 million units (Note: In a subsequent speech, Mr. Gao Feng, CMO of a leading domestic AI toy company, stated that the company's shipments accounted for about one-quarter of this total). Mr. Xing Xiaoci expects this number to exceed 10 million units by the end of this year.

From the trend of implementation, the integration of visual understanding through multimodality is the focus of Byte's layout. Mr. Wu Di, head of smart algorithms at Volcano Engine, provided two examples of hardware implementation: 1) Under the empowerment of large models, security cameras can not only serve as cameras but also act as housekeepers and personal assistants. 2) A desk lamp equipped with a camera can become a learning machine for solving problems, and when paired with a printer, it can help students organize their error collections.

Industry chain companies join the celebration, with Broadcom Integration, Starry Technology, Rokid, and others delivering keynote speeches.

The venue saw many representatives from Byte ecosystem companies, such as Zhongke Lanyun, Quectel, Starry Technology, and Broadcom Integration. Among them, Mr. Zhang Pengfei, Chairman of Broadcom Integration, and Mr. Chen Lijing, Vice President of Starry Technology, were invited to speak at the smart hardware sub-forum. Mr. Zhang focused on the advantages of low latency, high bandwidth, and ultra-low power consumption, highlighting how chip products optimize the AI experience on the Doubao ecosystem's edge; Mr. Chen focused on multimodal applications, stating that the company currently has chip products deployed in home, commercial, and wearable (such as glasses) sectors. Additionally, Mr. Wang Junjie, Vice President of the well-known domestic glasses brand Rokid, delivered a speech focusing on multimodal applications in glasses; Mr. Gao Feng, CMO of a leading domestic AI toy company, focused on end-to-end toy products in his speech, stating that the company is expected to release the world's first end-to-end AI toy within the year The release date of Xiaomi glasses is approaching, and the sentiment in the end-side sector is expected to improve.

According to the WeChat public account XR Vision, Xiaomi glasses are expected to hold a media day on June 16 and be released on June 26. The product is equipped with Qualcomm AR1 and Hengxuan 2700 chips, and features a Sony IMX 681 image sensor. Xiaomi is the first mobile phone manufacturer to release AI glasses. Considering the product's exposure and fan loyalty, there is a possibility that the subsequent sales tracking of this product may exceed expectations, and investors are advised to maintain high attention. Under the catalyst of this event, we are optimistic about the improvement of sentiment in the end-side sector represented by AI glasses. In addition to AI glasses in the Rayban META style, it is important to emphasize that investment in the glasses sector in 2025 should consider the evolution of product forms, from glasses without displays to single green display and full-color display glasses. Therefore, it is recommended to pay attention to the release rhythm of related products from manufacturers such as META, Rokid, and Yingmu; in the supply chain, attention should be paid to optical waveguides and MicroLED segments.

Risk factors:

Downstream demand may fall short of expectations; product innovation may stagnate; product application scenarios may be limited; rising costs of terminal hardware may impact demand; changes in the international industrial environment and intensified trade friction; the pace of AI commercialization may fall short of expectations; the iteration speed of large models may fall short of expectations