The spotlight in the AI industry has recently shifted to a red "little lobster"—OpenClaw. Industry professionals note that while the direction of AI development was anticipated, the speed and impact on productivity have surpassed expectations. MiniMax Vice President Yan Yijun remarked that various practitioners have actively reached out to explore or build similar systems.
The most responsive players are often those closest to revenue streams. Following the Lunar New Year holiday, domestic large language models have surged in usage rankings, driven by the OpenClaw trend, leading to significant revenue growth. Companies are collectively advancing agent projects, integrating "Claw" capabilities into their product portfolios.
According to Wu Wei, Partner at非凡资本, firms such as TENCENT, KNOWLEDGE ATLAS, and Kimi are pursuing similar paths for two key reasons: competing for the next-generation application gateway and addressing mounting pressure on pure model sales. In the current landscape, agent technology represents a rare direction that aligns long-term strategic vision with rapid, revenue-generating implementation.
Amid the excitement, security concerns have emerged. Frequent vulnerabilities in OpenClaw prompted the National Internet Emergency Center to issue risk warnings highlighting core threats like prompt injection, accidental deletion, and plugin poisoning. Capital markets reacted swiftly, with MiniMax and KNOWLEDGE ATLAS dropping over 6% on March 11.
The lobster phenomenon has not only opened a pathway for AI commercialization but also initiated a prolonged contest centered on gateways, ecosystems, and security. The industry consensus is that OpenClaw may not be the ultimate winner or final form, yet active participation by domestic firms is accumulating advantages for future iterations.
Domestic models have become a "traffic goldmine." Since late January 2026, OpenClaw's influence spread from Silicon Valley to China. Step 3.5 Flash explained that its popularity stems from addressing genuine user needs, transitioning agents from niche to general use and from developer communities to mainstream audiences.
Domestic model providers coincidentally released a wave of new products around the Lunar New Year. Data from OpenRouter indicates that from February onward, Chinese models consistently dominated the top five weekly rankings, accounting for more than half of the leading positions.
Understanding OpenClaw's business model is key: it is an extremely "voracious" consumer of tokens, currently ranking as the highest token-consuming application on the OpenRouter platform. Heavy users consume between 30 million and 100 million tokens daily, whereas typical ChatGPT users may only reach monthly consumption in the millions. This high token usage drives substantial API calls to cloud-based models, benefiting cost-effective domestic providers.
IDC China Research Director Lu Yanxia noted that widespread attempts to deploy OpenClaw and similar products are fueling token consumption, thereby boosting revenue for model and cloud providers. OpenRouter data shows that Chinese models—Step 3.5 Flash, Kimi K2.5, and MiniMax M2.5—hold the top three spots in daily and monthly rankings.
For model companies, this trend offers structural benefits: not only addressing urgent monetization needs but also gathering valuable high-quality task data through frequent calls by global developers. Yan Yijun projected that computing power and token usage could grow several-fold, with user preference shifting toward the most cost-effective models.
The next phase of competition revolves around "land grabs." As users grow accustomed to employing agents for tool invocation and task completion, these agents are poised to become the next super gateway, akin to search engines and app stores. No player wants to be left behind.
On March 10, KNOWLEDGE ATLAS launched AutoClaw, branded "澳龙," as China's first one-click local installation of OpenClaw. It features the optimized Pony-Alpha-2 model, over 50 pre-installed skills, and integration with tools like Lark. KNOWLEDGE ATLAS shares rose nearly 13% that day.
TENCENT also moved aggressively, introducing the full-scenario AI agent WorkBuddy on March 9. It fully兼容 OpenClaw skills without cloud deployment, connecting directly to WeChat Work for remote operation. Initial public testing faced login and service instability due to traffic surges. TENCENT shares increased over 7% on March 10, lifting its market cap back to HK$5 trillion.
TENCENT Chairman and CEO Ma Huateng noted in a social media post that a range of products—from self-developed to enterprise-grade lobsters—are in the pipeline. On March 11, Huawei Terminal BG CEO He Gang unveiled XiaoYi Claw, based on the HarmonyOS, currently in beta. It assists with document editing, PPT creation, and email responses, supporting multi-device coordination and preset personas. Huawei's open platform had already added an OpenClaw mode.
Incomplete statistics show a proliferation of similar offerings, including Alibaba's CoPaw, Volcano Engine's ArkClaw, Moonshot AI's KimiClaw, and MiniMax's MaxClaw, marking an "all-in Claw" moment for the AI sector.
Yan Yijun revealed that MaxClaw required four capacity expansions within 120 hours of launch, with continuous scaling ongoing. The enthusiasm echoes the fervor seen during last year's DeepSeek boom.
Industry observers view OpenClaw's current influence as a critical window. Sullivan China Director Li Qing described it as a battle for gateways and ecosystem positioning, where monetization is an outcome rather than the initial goal, with the deeper logic lying in building moats.
Domestic "lobsters" offer differentiated advantages. They lower deployment barriers, promoting AI accessibility through user-friendly, out-of-the-box solutions. Additionally, ecosystem integration enhances practicality: TENCENT's WorkBuddy connects with WeChat Work and QQ, while Huawei's XiaoYi leverages HarmonyOS's cross-device capabilities, embedding user habits within proprietary environments.
However, amidst the rush, underlying concerns persist. Security vulnerabilities and privacy risks from local deployments remain critical challenges. The National Information Security Vulnerability Database recorded 82 OpenClaw vulnerabilities from January to March 9, 2026, including 12 critical and 21 high-risk issues, spanning access control errors and code problems.
The National Internet Emergency Center's March 10 warning highlighted OpenClaw's weak default security settings, identifying four core risks: prompt injection, accidental deletion, plugin poisoning, and data leakage in sectors like finance and energy.
Wu Wei cautioned that current market attention to OpenClaw security remains insufficient, with many focusing only on surface issues like key leakage. The real challenges involve defining permission boundaries, controlling tool invocation, preventing memory pollution, establishing误触发 mechanisms, and building audit systems. He believes while OpenClaw may gain traction in personal or light team use, enterprise core integration will be a lengthy process.
Despite security concerns, the trend benefits the long-term development of the agent industry. Exposing security issues prompts the industry to establish safeguards, with domestic providers already addressing these aspects. For instance, KNOWLEDGE ATLAS representatives stated that AutoClaw includes preemptive measures against public network and port exposure, with restrictions and alerts for critical operations.
Citi analysis noted that TENCENT's WorkBuddy enhances usability and security through the CodeBuddy framework, incorporating unified accounts, billing, and enterprise-grade management and auditing.
Simultaneously, the trend is educating users and enterprises about agent capabilities, fostering token payment habits, and providing sustainable support for model and cloud providers. The ecosystem continues to thrive: Step 3.5 Flash reported rapid growth in OpenClaw-related software and communities, driving demand for foundational models and computing power. Model inference resources are again facing shortages.
The prevailing view is that OpenClaw is unlikely to be the final victor or form, with multiple iterations expected. Even if the current hype subsides, domestic firms' active involvement holds value. Wu Wei indicated that major players will enhance infrastructure around permissions, security, memory, workflows, and multi-platform collaboration, advancing agents from demonstrative demos to deliverable products.
He observed a shift in focus during recent OpenClaw discussions toward deployment, secure operation, and business integration, signaling the industry's progression toward practical implementation.
Comments