Kaiyuan Securities released a research report stating that OpenClaw experienced explosive growth in February 2026, topping the GitHub global trending chart and becoming the most popular open-source project. Since February, leading domestic manufacturers like Zhipu and Tencent have intensively launched Claw-like products or solutions based on the open-source OpenClaw code. The application of Agent AI is driving a sharp increase in Token consumption. According to IDC forecasts, by 2030, the number of active global AI agents will reach 2.216 billion, and annual Token consumption is projected to surge from 0.0005 Peta Tokens in 2025 to 152,000 Peta Tokens, representing growth exceeding 300 million times. It is recommended to focus on the AI inference computing power industry chain (including chips, complete machines, liquid cooling, power supplies, and other segments) and the edge AI industry chain. The main views of Kaiyuan Securities are as follows:
The OpenClaw trend continues, with major domestic companies intensively releasing Agent AI products similar to Claw. OpenClaw is an open-source AI agent framework. Its initial version, Clawdbot, was released on the GitHub community on November 24, 2025, and was later renamed OpenClaw. In February 2026, OpenClaw saw explosive growth, topping the GitHub global trending chart and becoming the most popular open-source project. Compared to other Agent products, OpenClaw features local-first operation, autonomous tool invocation, cross-application execution, multi-IM integration, and continuous online operation, naturally possessing "digital employee" attributes that better align with users' real AI needs, leading to widespread popularity in China. Since February 2026, leading domestic manufacturers including Zhipu, Tencent, Huawei, Alibaba, ByteDance, Kimi, MiniMax, and Xiaomi have intensively launched Claw-like products or solutions based on the open-source OpenClaw code. Major domestic cloud service providers like Tencent Cloud, Alibaba Cloud, and Volcano Engine have also launched OpenClaw deployment services on their cloud platforms. Zhipu's AutoClaw is the first domestic "one-click installation" local version of OpenClaw, featuring the Pony-Alpha-2 model optimized specifically for OpenClaw, pre-installed with over 50 mainstream skills, and supporting integration with instant messaging tools like Feishu. Tencent's WorkBuddy is deeply compatible with all OpenClaw skills and requires no cloud deployment; it can be operated remotely after download by connecting to WeCom. Kaiyuan Securities believes that OpenClaw may mark the transition of large AI models from the "dialogue era" to the "agent era."
Agent AI applications drive a surge in Token consumption, potentially leading to exponential growth in inference computing demand. Agent AI features machine-autonomous invocation, long-context memory, multi-toolchain collaboration, and multi-agent concurrency. These advancements enhance the AI interaction experience while shifting the demand on the inference side away from being primarily driven by human interaction frequency. This "plan-execute-feedback-replan" workflow causes Token consumption to leap from previous linear growth to exponential growth. Taking OpenClaw as an example, users only need to set a goal, and it can autonomously break down tasks and invoke models/tools/interfaces to complete the entire process. A single task can easily consume hundreds of thousands to millions of Tokens. According to the National Data Bureau, China's daily Token consumption was only 100 billion in early 2024, but by the end of June 2025, daily Token consumption had exceeded 30 trillion, a more than 300-fold increase in a year and a half. IDC forecasts that by 2030, the number of active global AI agents will reach 2.216 billion, and annual Token consumption is projected to surge from 0.0005 Peta Tokens in 2025 to 152,000 Peta Tokens, representing growth exceeding 300 million times.
OpenClaw's continuous penetration into PCs, phones, wearables, and other devices suggests edge AI may undergo a qualitative transformation. Before the release of OpenClaw, traditional agents suffered from limitations such as lack of cross-device scenario versatility, memory loss, limited context windows, absence of proactive execution mechanisms, and poor autonomous scheduling capabilities. OpenClaw adopts a modular architecture of Gateway + Agent + Skills + Memory, featuring complete session management and a memory system message routing. Its "heartbeat" mechanism periodically reads task lists in the workspace, enabling autonomous control of software like browsers, email clients, and chat tools, as well as reading/writing files, running scripts, and managing schedules without human intervention. This signifies AI transitioning from a conversational role to an "executor" role, with the potential to evolve into more intelligent personal AI assistants and digital employees in the future. Beyond PCs, manufacturers of mobile phones and wearable products are also actively adapting to OpenClaw-like agents. On March 6, Xiaomi's miclaw began a limited beta test. miclaw runs as a system application, encapsulating the phone's system capabilities into over 50 system tools and ecosystem services, and building a large model-based inference-execution engine. On March 11, Huawei disclosed that its HarmonyOS-based XiaoYi Claw is already in beta version. XiaoYi Claw can assist users with tasks like document editing, creating PPTs, and auto-replying to emails, supports multi-device collaboration, and comes with multiple preset personalities. On March 14, the head of the OpenClaw community announced on platform X that an OpenClaw version for smart glasses would be developed based on the Rokid Glasses developer tools. Kaiyuan Securities believes that OpenClaw promotes the shift of edge AI from "reasoning" to "execution," redefining the capability boundaries of edge AI and is expected to catalyze more edge AI application scenarios and terminal categories.
Recommended investment targets include: Hygon Information (688041.SH), Lingyi Intelligent (002600.SZ), Dongshan Precision (002384.SZ), Oulute (300870.SZ), Aohai Technology (002993.SZ), Transsion Holdings (688036.SH), etc. Beneficiary targets include: Cambricon (688256.SH), Zhiwei Intelligent (001339.SZ), Hesheng New Materials (002290.SZ), Industrial Fulian (601138.SH), Luxshare Precision (002475.SZ), Universal Scientific Industrial (601231.SH), Hongrida (301285.SZ), Speedy (300322.SZ), Suzhou Tianmai (301626.SZ), SICC (02631), etc.
Risk warnings include risks related to AI progress falling short of expectations, international trade environment risks, and risks of intensified industry competition.
Comments