A report from Guolian Minsheng Securities indicates that the "Lobster Wave" ignited by OpenClaw signifies that this open-source, local-first AI Agent framework, capable of terminal keyboard/mouse takeover and autonomous execution of complex tasks, is systematically restructuring the supply and demand dynamics of cloud computing. The business logic for cloud providers has evolved from simple "underlying computing power leasing" to becoming "workstation providers for Agent-based digital employees." By capturing application entry points through one-click deployment and price subsidies, cloud providers are progressively gaining control over the underlying distribution rights for next-generation AI applications.
The primary viewpoints of Guolian Minsheng Securities are as follows:
**Event Overview:** On March 8, 2026, Tencent officially opened QQ bot access to OpenClaw for individual users. After logging in via mobile QQ scan, users can create a bot with a single click, with a maximum of five bots per QQ account. Following creation, binding with OpenClaw is quickly completed using three preset instructions, supporting multimedia interactions including Markdown, images, voice, and files. Concurrently, Xiaomi announced a limited beta test for its mobile Agent product Xiaomimi claw, effectively shifting the computing power competition from the cloud directly to mobile devices and smart terminals within the "human-vehicle-home" ecosystem.
**Industry Evolution Trend: Computing Power Inventory Reduction and Entry Point Seizure** Execution hubs like OpenClaw are transforming AI computing power consumption from "Chatbot-style dialog requests" to "round-the-clock continuous operation." Compared to single-turn conversations, Agent tasks involve significantly higher demands for tokens, web searches, long-term memory, state storage, and persistent connections, leading to a marked increase in per-user cloud resource consumption. This shift brings two key changes:
1. **Exponential Growth in Underlying Computing Power Demand:** Promotions by cloud providers and community deployment cases show that as OpenClaw gains rapid popularity domestically, lightweight application servers (e.g., 2-core/2GB, 2-core/4GB) have become one of the most common deployment specifications, significantly increasing cloud deployment demand. This new demand wave is particularly effective in utilizing lower-end CPU instances, basic storage, and bandwidth resources. 2. **Supply-Side "One-Click Deployment" and Entry Point Capture:** Major domestic cloud providers have swiftly bypassed complex Docker or command-line deployments by launching dedicated entry points for lightweight servers pre-installed with OpenClaw images. For instance, Tencent Cloud has introduced notable "new user acquisition" incentives around its intelligent agent development platform, including "Free Direct Claim," "15,000 PU/month resource points," and "New users receive a 1-month free tier package." Furthermore, Tencent Cloud's developer community disclosed offline OpenClaw events attracting "nearly a thousand registrants," indicating a dual-pronged strategy using pricing advantages and ecosystem activities to drive Agent adoption, aiming to convert short-term high traffic into long-term SaaS subscription retention.
Guolian Minsheng Securities believes this strategic positioning is fundamentally aimed at deeply locking user application data, API keys, and network environments within the provider's own cloud ecosystem. Therefore, the current competition among cloud providers surrounding OpenClaw is intended to "capture the initial deployment location for Agents." Once a user completes the first deployment on a specific cloud platform, that provider simultaneously gains control over default models, default skills, default knowledge bases, default billing, and subsequent scaling entry points. Tencent Cloud's official activity page indicates OpenClaw can be deployed on its ADP platform within 3 minutes, with the system automatically purchasing servers, initializing the environment, and pre-configuring default models and skills. Its free trial package offers 15,000 PU/month, while platform billing documentation explicitly states "model calls are billed separately." This indicates Tencent has shifted OpenClaw from a pure "model call" object to a dual-layer charging model of "platform service + model calls." Moreover, Tencent Cloud switched its old token resource package system to a subscription-based plan in January 2026, suggesting that Agent scenario pricing units are transitioning from单纯的 tokens to a composite billing model of "package resource bundles + deployment fees + model calls."
Guolian Minsheng Securities contends that the shared logic among cloud providers is clear: first migrate users from local installations to the cloud using subsidies and one-click deployment, then achieve subsequent ARPU growth through model calls, plugins, searches, knowledge bases, long-term memory, and multi-channel access.
**Industrial Chain Value Distribution:** Amid the confirmed Agent "gold rush," cloud providers, acting as "shovel sellers," are poised for a triple value reassessment:
1. **High Certainty: Large-Scale Monetization of Basic Instances:** OpenClaw has fundamentally blurred the lines between personal PCs and cloud servers, converting long-tail, low-end servers into "cloud workstations" for Agents. This is expected to drive increased demand for cloud providers' edge computing nodes, reduce inventory of lightweight application servers, and enhance recurring rental revenue. On the hardware side, shipments of domestic AI computing infrastructure are anticipated to receive substantial support. 2. **High Elasticity: Non-linear Growth in Token Calls:** When executing end-to-end complex tasks, Agents require frequent information retrieval, logical reasoning, and multi-step planning, leading to token consumption per task that is tens or even hundreds of times greater than traditional Q&A. This drives exponential growth in API billing revenue for the model layer, providing a dual boost in business volume and data feedback for model vendors tied to major cloud platforms, such as MiniMax and Zhipu. Essentially, cloud providers are transforming from "underwriters of single models" into "billing layers for multiple models." For example, Alibaba Cloud's Bailian CodingPlan supports Qwen, Kimi, MiniMax, and GLM simultaneously; Tencent Cloud's ADP free package deeply integrates GLM5, MiniMax2.5, Kimi2.5, and DeepSeek. While model vendors develop their own products, they are also redistributed on platforms like Alibaba Cloud and Tencent Cloud. When model calls, knowledge bases, plugins, and deployments are all settled through the cloud provider's own platform, the value of providers offering a "model supermarket + billing platform + managed operations" bundle typically exceeds that of pure IaaS vendors selling only underlying resources. 3. **Highest Long-Term Barrier: Premium for AI Platforms and Full-Chain Tools:** The enterprise deployment of OpenClaw crosses three major thresholds: deep integration of enterprise knowledge, secure sandbox permission control, and business system API connectivity. Cloud providers can widen their profit margin compared to pure computing power distribution by offering high-margin customized AI work platforms, enterprise-grade security networks, and persistent storage. As Agents move into enterprise scenarios, model intelligence is not the sole consideration; skill management, permission management, auditing, stability, and operational governance are equally critical. Tencent Cloud has even decoupled its ADP agents from OpenClaw agents via plugins for high-permission complex tasks like browser automation. Platformization, toolification, and governance are likely OpenClaw's most profound long-term impacts on cloud providers, as enterprises ultimately purchase "controllable Agent platforms."
**"Out-of-the-Box Products" Further Reinforce Cloud Infrastructure** As technical barriers lower, major model vendors like Kimi and MiniMax, hardware terminal giants like Xiaomi, and small-to-medium enterprises are successively launching out-of-the-box variants of OpenClaw. This trend will strongly reinforce public cloud infrastructure in two dimensions:
1. **Major Model/Terminal Vendors: Direct Amplifiers of Computing Power Demand:** Products like KimiClaw and MaxClaw from major model vendors, or Xiaomimi claw from Xiaomi, essentially alleviate users' local hardware concerns, enabling out-of-the-box use without server purchases. This consumer-friendly, simplified experience not only locks in significant model API consumption from a large user base but also directly channels the resulting high concurrent requests and contextual memory storage demands to their underlying public cloud providers. As top clients of cloud service providers, the prosperity of these vendors' end-side applications will directly increase the load peaks and network expansion needs of the underlying cloud infrastructure. 2. **Small-to-Medium Enterprises: Computing Power Distributors for Vertical Scenarios:** Companies like Shunwang Technology, with deep expertise in specific terminal scenarios such as internet cafes, esports, and cloud gaming, and capabilities in computing cloud and cloud terminals, are well-positioned to act as computing power distributors and middleware integrators for vertical scenarios during Agent implementation. In this ecosystem, SMEs serve as "computing power distributors" and "middleware," purchasing base resources in bulk from major cloud providers and leveraging their own edge computing networks to deliver high-threshold AI capabilities to non-technical users. This not only directly consumes public cloud elastic computing and object storage resources but also incorporates hard-to-reach,下沉 traffic into the overall cloud computing billing base through their terminal scenarios.
**Investment Recommendations:** Focus on comprehensive cloud providers possessing capabilities in computing power, model/API access, and platform tools, such as Alibaba Group Holding Limited (09988), Tencent Holdings Limited (00700), Baidu, Inc. (09888), and Kingsoft Cloud Holdings Limited (03896). Also consider companies with cloud resources and AI operational foundations, such as Wangsu Science & Technology Co., Ltd. (300017.SZ), UCloud Technology Co., Ltd. (688158.SH), Beijing Capital Online Data Service Co., Ltd. (300846.SZ), Dongyangguang (600673.SH), QingCloud Technologies (688316.SH), Tongniu Information (300895.SZ), and Hongjing Technology (301396.SZ).
**Risk Warning:** Risks include profit mismatch due to rigid growth in capital expenditures and insufficient revenue elasticity from AI, and the risk of being replaced as core clients build their own computing power platforms.
Comments