Guolian Minsheng Securities: Who benefits from Token "inflation"? Three main trends in core assets worth paying attention to.

Guolian Minsheng Securities released a research report suggesting to focus on the main line of "Agent penetration rate improvement - Token inflation - cloud and model repricing". In the short term, it is recommended to pay priority attention to the cloud platform, as its business model is clear, quick to implement, and profit improvement is expected to be realized more quickly; in the medium term, the focus should be on model manufacturers, who may be able to handle higher frequency calls at lower costs and achieve growth in revenue from increased flow and subscription; in the long term, domestic AI computing power chips are favored, with the expansion of inference demand and the trend towards self-controllable autonomy, domestic chip manufacturers are expected to have stronger performance resilience. Overall, it is recommended to focus on the core assets of the three main lines of "chip", "model", and "cloud". The main views of Guolian Minsheng Securities are as follows: From the chatbot Siasun Robot&Automation to the "proactive" Agent intelligent body. The emergence of OpenClaw represents the transition of Agent from "just answering" to "proactively executing". Its core value lies in enabling AI to be online 24/7, autonomously decompose tasks, call tools, read and write files, cross-platform scheduling, and closed-loop delivery. AI is moving from a conversation assistant to a truly implementable "digital employee". The bank expects that this means the billing logic of the internet and enterprise IT is shifting from past traffic, time CKH HOLDINGS entrance competition to Token consumption, completion of execution, security governance, audit tracking, and resource supply billing. "Dragon Lobster" fiery inspiration indicates the direction of benefiting from the "inflation" of Token demand. The fiery emergence of "Dragon Lobster" predicts the nonlinear growth of Token demand in the Agent era, which may directly bring unexpected AI computing power demand and the calling demand of large models. Meanwhile, cloud providers are no longer just selling computing power, but are also competing for the deployment entrance, model distribution rights, and subsequent expansion rights of Agents. The bank predicts that the continuous operation of Agents will significantly increase Token consumption, internet searches, state storage, long connections, and multi-step reasoning demand, thereby driving a systematic reassessment of chips, models, and cloud platforms on three levels. "Chip": Cambricon, Hygon Information Technology, YunTianlifly and other domestic AI chips. The computing power demand in the Agent era is not only GPU inference, but also includes tool calls, environment construction, task scheduling, long-term memory, and high-concurrency CPU collaboration. The importance of computing power base has been significantly increased. Domestic AI chip manufacturers such as Cambricon, on the one hand, benefit from the expansion of inference demand brought about by Token inflation, and on the other hand, also benefit from the trend of domestic substitution and self-controllable autonomy in the industrial chain. "Model": MiniMax, KNOWLEDGE ATLAS. The ability to code large models in the Agent era shows a nonlinear growth, and its comprehensive requirements for unit cost, long text, multi-step reasoning, programming capabilities, etc., make the advantages of domestic large models evident. MiniMax's advantages lie in its cost-effectiveness, long context, and programming/visual execution capabilities, making it more suitable for high-frequency workflow calls; KNOWLEDGE ATLAS has world-leading coding technology capabilities, and the price increase reflects the transition of models from price wars to tiered subscriptions and pricing increases. "Cloud": KINGSOFT CLOUD, Wangsu Science & Technology, UCloud, etc. The cloud is the high-certainty "shovel seller" in this round. KINGSOFT CLOUD, UCloud, etc., benefit from one-click deployment, model hosting, knowledge base, Agent platform subscription, and enterprise-level governance requirements; Wangsu Science & Technology benefits from the uplift in machine traffic, CDN, API gateway, edge security, and value-added traffic governance. Overall, the cloud layer has high certainty, the model layer has great flexibility, and the chip layer is the long-term foundation. Risk warning: Agent landing pace is slower than expected; product conversion rate of model manufacturers is lower than expected; model price war; disturbance in computing power supply; deviation in elasticity assessment of CPU demand.