Hualong Securities: Volcano Engine FORCE Conference Held to Accelerate Model Implementation to Unlock Computing Power Demand Potential

The Zhitong Finance App learned that Hualong Securities released a research report saying that the Volcano Engine FORCE conference was held recently, a number of major domestic models were officially launched, and the AI industry trend is gradually becoming clear. The bank believes that the inter-partition pricing model fully takes into account the demand distribution of enterprises in actual application scenarios, can reduce the overall cost of customers, and the cost performance advantage of the model continues to improve; newly upgraded agent development products help enterprises implement agents in production scenarios, while also unleashing the demand potential for inference computing power, and the bank maintains a “recommended” rating for the computer industry.

Recommended attention: (1) AI applications: Zhuoyi Information (688258.SH), Dingjie Digital Intelligence (300378.SZ), Caixun Shares (300634.SZ), Tournament Information (300687.SZ), Jinqiao Information (603918.SH); (2) Computing power: Shenzhen Trust (300454.SZ), Capital Online (300846.SZ), Hongjing Technology (301396.SZ), Yunsai Smart Link (600602.SH).

The main views of Hualong Securities are as follows:

Incidents:

On June 11, 2025, the Volcano Engine FORCE conference was held. At the conference, Volcano Engine officially released the Doubao Big Model 1.6, the Douba· Video Generation Model Seedance1.0Pro, and the Douba· Voice Podcast Model.

Opinions:

The Doubao Big Model 1.6 was officially released, and the overall price was 63% lower than the DeepSeek R1.

Doubao Big Model 1.6 pioneered the “input length” pricing model. Within the 0-32K input range used by the vast majority of enterprises, Doubao Big Model 1.6 is priced at 0.8 yuan/million tokens (input) and 8 yuan/million tokens (output), which is the lowest price for all input ranges. Because the input length of most tasks is within 32K, and the input/output ratio is 3:1, Doubao Big Model 1.6 is more cost-effective in enterprise usage scenarios, and the overall cost is 63% lower than DeepSeek R1. The bank believes that the inter-partition pricing model fully takes into account the distribution of enterprise demand in actual application scenarios and can reduce the overall cost of customers, thereby improving the cost performance advantage of the model. Other leading vendors may follow up on inter-partition pricing, thereby further reducing the average cost of downstream call models and increasing call demand.

Newly upgraded agents develop products to help enterprises implement agents in production scenarios. The conference built 5 new and upgraded products for enterprise agents. In the field of AI-assisted programming, since the launch of the AI IDE product TRAE in January 2025, there have been more than 1 million monthly active users. For development scenarios, TRAE provides functions such as code completion and partial code generation, and supports natural language dialogue to complete the entire development process. The next version will integrate different agents and tools. On the MCP side, a new Volcano Engine MCP service was released, which can link the agent development system, large model tool ecosystem, and cloud services. It is connected to TRAE, Volcano Ark, and buttons, and supports the selection of 200+ MCP services. The bank believes that currently AI production tool products, represented by AI IDE products, are unleashing demand potential. Meanwhile, major manufacturers promoting MCP services and product iterations to support agent development will continue to clear the path of agent implementation, which is a necessary condition for the agent ecosystem to prosper.

Tokens usage has increased dramatically, and demand potential is being unleashed. By the end of May 2025, the average daily token usage of the Doubao Big Model was over 16.4 trillion dollars, an increase of 137 times compared to when the Doubao Big Model was first released in May 2024. It has only been more than a year since the release of the Doubao Big Model, and tokens have achieved explosive growth, reflecting that the model's reasoning ability is being accepted and recognized. After Deepseek was released, the big model industry focused more on inference and cost performance in competition and catch-up. The economies of scale brought about by “model equalization” are showing, while further boosting computing power consumption.

Risk warning: AI technology innovation falls short of expectations; policies fall short of expectations; geopolitical risks; AI regulatory risks; macroeconomics fall short of expectations.