Guolian Minsheng Securities: "Western Shrimp Rearing in the East" reflects the ecological dislocation of the AI application landing wave. Significant cost advantages of domestic large models.

Guolian Minsheng Securities released a research report, stating that the current "lobster craze" apparent phenomenon in China reflects the "FOMO-style anxiety" of domestic large factories for the OpenClaw market, but actually reflects the ecological mismatch between domestic and overseas in the wave of AI application landing in the Agent era. Compared to overseas, China has cost advantages, traffic advantages, and policy advantages in this wave of "lobster craze". From a cost perspective, driven by OpenClaw, the consumption of domestic model tokens is accelerating, with OpenRouter data showing that the consumption of large models in recent times has reached a new high, with a steep growth rate, and China's model token consumption ranking first. The rapid collision of token consumption is also evidence of cost advantages. Guolian Minsheng Securities main points are as follows: Event: On March 10th, KNOWLEDGE ATLAS officially launched AutoClaw (Chinese name: Oulong)a locally installable version of OpenClaw with full native capabilities. Currently, KNOWLEDGE ATLAS provides a certain amount of free quota for users to experience the capabilities of AutoClaw. KNOWLEDGE ATLAS introduced that AutoClaw Oulong comes pre-installed with over 50 mainstream Skills, ready to use, covering high-frequency scenarios such as content creation, office work, coding, marketing, financial research, and supporting one-click access to instant messaging tools such as Feishu. It can interface with any model's CodingPlan/API, including APIs from model manufacturers such as DeepSeek, Kimi, MiniMax, and GLM. In fact, in addition to KNOWLEDGE ATLAS, many domestic large factories have actively responded to this wave of "lobster craze" and participated in the localization process of OpenClaw. Since February, companies like KNOWLEDGE ATLAS, MiniMax, and other large model companies and major cloud factories have successively released related Claw products. By comparing the product features of OpenClaw and domestic entities, there are currently differences in deployment methods and model sources. From the perspective of deployment methods, the participation of various entities in this wave of "lobster craze" presents a "variety of ways" situation. Unlike OpenClaw's emphasis on local deployment, domestic large factories' "lobster products" have three deployment methods: local deployment, cloud deployment, and hybrid deployment, providing more options. As an open-source AI intelligent body, OpenClaw was designed to emphasize localization and flexibility. According to its founder Peter Steinberger, the safest and most flexible way to use OpenClaw is to deploy models locally + OpenClaw + integrate with existing instant messaging tools. Guolian Minsheng Securities believes that this phenomenon may mainly be due to the domestic "lobster products" in this round paying more attention to the ease of deployment and security. In terms of model sources, most entities in this round are currently focused on multi-model matching, but some companies choose to rely solely on their own models. In comparison, OpenClaw supports a wider variety of large models, involving mainstream models both domestically and overseas; while China's "lobster products" mainly focus on supporting the call of domestic mainstream large models, with some currently only relying on their own models as the main source. This may be due not only to the fact that the dominant large factories in this round of "lobster products" are more concerned about the cost and operability of calling domestic large models, but also the competition for the "new distribution rights" in the Agent era. Why does this round of "lobster craze" sparked by Openclaw exhibit the feature of "growing lobster in the East"? Although the rise of OpenClaw products in this round came from overseas, since February, the level of attention domestically has rapidly increased, with many participants entering the market, showing the feature of "growing lobster in the East". Combining the above comparison, this apparent phenomenon of "growing lobster in the East" is the domestic large factories' "FOMO-style anxiety" about the OpenClaw market, but actually reflects the ecological mismatch between China and overseas in the wave of AI application landing in the Agent era. Compared to overseas, China has cost advantages, traffic advantages, and policy advantages in this round of "lobster craze". From a cost perspective, the domestic large model ecosystem has formed a unique "low-priced API". In terms of calling prices, the API calling price for domestic models is roughly 1/6 of similar products overseas. This advantage comes from cheaper electricity behind domestic computing power, more flexible hardware configurations, and price competition among model manufacturers. Driven by OpenClaw, the consumption of domestic model tokens is accelerating, with OpenRouter data indicating that the consumption of large models has reached a new high, with a steep growth rate, and China's model token consumption ranking first. The rapid collision of token consumption is also evidence of cost advantages. From a traffic perspective, this round of "lobster craze" cannot be separated from the active participation of large factories. For large factories, deploying OpenClaw may not only be related to revenue growth in the Agent business itself but also includes the competition for the potential "new distribution rights" in the Agent era. Looking ahead, with the continued deployment of Agents, there may be a new restructuring of the interaction between humans and the digital worldwhen the operating subject transitions from humans to Agents, new traffic entry points may also shift. Although this round of OpenClaw may still face technical immaturity, its reversal from AI "Chat" to "Work" could lead to market concerns about changes in new platform entry points, thereby leading to widespread "FOMO" and active deployment by large factories in this round. In addition, large factories represented by Tencent may further trigger attention in the market with their social fission effect, driving the spread of attention to the "lobster craze". From a policy perspective, local governments in China have recently attached great importance to OpenClaw, which has had a catalytic effect on sentiment. Shenzhen and Wuxi have taken the lead in issuing special policies, becoming pioneers of the "lobster craze". 1) On March 8th, according to reports from Shenzhen, the Longgang District of Shenzhen recently solicited opinions on the "Measures to support the development of OpenClaw & OPC (draft for comments)". It proposes to encourage market-oriented and professional platforms to launch the "lobster service area", provide free deployment services for OpenClaw, and give certain subsidies to eligible entities. 2) On March 9th, Wuxi High-tech Zone issued the "Measures to support the integration and development of OpenClaw and other open-source community projects with the OPC community (draft for comments)" released. The 12 "lobster farming" policies, from basic support to industrial landing, from talent introduction to security and compliance, release a package of substantial incentives, with the highest single support reaching up to 5 million RMB. At the deployment level, local cloud platforms providing free deployment and development toolkits will receive full subsidies of up to 1 million RMB. In terms of investment targets, it is recommended to focus on: 1) MINIMAX-WP(00100) and KNOWLEDGE ATLAS(02513), which have model or application capabilities and benefit from the potential accelerated growth in API call-related revenue; 2) BABA-W(09988) and TENCENT(00700), Internet platforms with computing resources, model capabilities, and coordinated application scenarios; 3) KINGSOFT CLOUD(03896), BIDU-SW(09998), and KINGDEE INT'L(00268), which have significant computing power and token consumption and are expected to drive related cloud service demand; 4) ILUVATAR COREX(09903), BIREN TECH(06082), HUA HONG SEMI(01347), and Montage Technology(06809), with strong deterministic AI computing hardware layers; 5) US stocks such as SNDK.US, MU.US, NVDA.US, INTC.US, LITE.US, and CMCM.US. Risk Warning: AI model technology development may fall short of expectations; commercialization of AI applications may not progress as expected; deterioration in the competitive landscape of the AI industry, etc.