Zhongjin: Technology reduces costs, scenarios elevate, AI companions provide deep emotional value.

CICC released a research report stating that the wave of AI big models is expected to bring changes to content production, distribution methods, and networks. User and AI roles interact bidirectionally, traditional social network effects are gradually disintegrating, and AI becomes the center of the network. An "TikTok" era of AI may appear, but it still faces multiple challenges such as technology, user retention, commercialization, and computing power costs. AI companion products currently lack product depth, gameplay is still in its early stages, and future gameplay innovation may be one of the key factors in long-term competition in this category. The current main extension direction is to use hardware as a carrier, focus on high-demand vertical scenes such as education and games, and expand the target audience to the elderly and children. CICC's main points are as follows: AI companion applications provide emotional and practical value, and the core barrier is product power. The core elements of AI companion applications are anthropomorphism, personalization, real-time interaction, immersion, and cultivation, meeting users' needs for companionship, entertainment, fantasy, and efficiency enhancement, with users mainly concentrated in the younger demographic. AI applications land and develop rapidly in the companion scenario, and the firm believes that this is mainly due to strong user demand, low technical barriers, and a high fault tolerance rate in the track itself. The key to landing is product power, and Chinese companies have advantages in engineering talent and product refinement capabilities in terms of product power. Technology: Cost reduction helps with mass production, multimodal enhances product experience. The Mixed Expert Model (MoE) reduces costs and empowers AI dialogue scenarios. Long-text capabilities are gradually becoming industry consensus, and linear attention models may be a better implementation path. In terms of products, the firm believes that multimodal enriches interactive forms, with mature routes for image generation, audio generation, and video generation. End-to-end multimodal is expected to open up new efficiency and performance spaces. The firm believes that the ability of AI companion applications to generate multimodal and personalized materials based on interactive content and user preferences is an important advantage compared to other entertainment applications, and multimodal capabilities are expected to improve user long-term retention rates and willingness to pay. Applications: Combining technology and warmth, following product and gamification design concepts. Replika focuses on emotional companionship and psychological counseling, Character.AI shapes products from a general chatbot perspective, MiniMax's Talkie and Xingye build AI interactive content platforms, Zhu Meng Island focuses on novel scenarios, ByteDance's AnDoor and CatBox emphasize social attributes, and Natural Selection EVE positions as a 3D AI companion. Risks Model iteration effects fall short of expectations, industry competition intensifies, and commercialization does not meet expectations.