"Lobster" concept stocks retreat; experts advise ordinary users not to blindly "raise lobsters"

China Economic Journalists Gu Mengxuan and Li Zhenghao Report from Guangzhou and Beijing

As the open-source AI agent “OpenClaw” (whose icon features a red lobster, nicknamed “Lobster” by netizens) becomes popular online, a new trend in the AI community has emerged—“raising lobsters.” The popularity of this “lobster” is like a digital storm sweeping across the internet. Netizens not only regard it as the “strongest worker,” but also have spawned a “lobster-raising” industry chain, turning “Have you raised a lobster?” into a new social greeting.

As the “raising lobsters” craze continues to grow, various issues have gradually surfaced, with security risks and usage costs being the most prominent concerns. The Ministry of Industry and Information Technology, the National Internet Emergency Center, and other authoritative agencies have issued multiple risk alerts, urging users to exercise caution.

The capital market has also responded. Although from February 1 to March 10, “lobster” concept stocks saw notable gains, on March 11, related stocks experienced declines.

Zhang Xinyuan, head of research at Kefangde Think Tank, told China Business Journal that the short-term explosion of OpenClaw is mainly because it transforms AI from “only capable of conversation” to “able to directly control computers and perform tasks,” precisely targeting automation in office work. Coupled with its open-source, free, locally deployable, and easy-to-use features, it quickly gained traction within communities and went viral.

Compared to traditional large models, Zhang Xinyuan pointed out that the advantage of “lobster” lies in its ability to perform tasks and control systems, automating workflows across software; supporting local deployment to keep data off the cloud and enhance privacy; being compatible with multiple models and lower costs; and continuously expanding capabilities through plugins, upgrading from a “Q&A tool” to an “automation assistant.”

“Digital Employees” Are Not Cheap

Relevant data shows that OpenClaw is an open-source autonomous AI virtual assistant software project developed by software engineer Peter Stenberg. It was first released on GitHub at the end of 2025 under the name Clawdbot, later renamed Moltbot, and finally settled on the current name.

In early 2026, this AI project gained attention for its ability to autonomously handle complex tasks within applications and online services based on user instructions.

A research report from Huaxi Securities’ computer team noted that OpenClaw stores configuration data and interaction history locally, giving it a more persistent memory. Driven by natural language commands, it can perform file operations, process automation, browser automation, and multi-IM platform interactions locally or in private cloud environments, achieving a leap from “dialogue-based suggestions” to “automated execution.” It is a self-hosted AI digital employee aimed at individuals and enterprises.

Regarding costs, Guosheng Securities pointed out that fundamentally, OpenClaw is a “Token consumer.” Its research report states that OpenClaw causes token consumption to increase more than fourfold in a month, creating a rigid demand for computing power.

Fund researcher Bi Mengran from Gushang Fund told reporters that previously, AI token business mainly focused on the B2B sector, with enterprise clients calling large models via APIs, resulting in limited token consumption and low per-transaction value. The explosion of “lobster” has allowed C-end users to participate widely, with ordinary users deploying in the cloud and using plugins, leading to high-frequency token consumption and directly boosting revenue for large model vendors.

Regarding installation costs, reporters learned that recently, Tencent Cloud Lighthouse engineers have been providing free one-stop installation tutorials for OpenClaw, and there are also “Lobster/OpenClaw on-site installation” services priced between 300 and 1,000 yuan, with 500 yuan per session being most common. Services include local deployment, debugging, and basic usage guidance, while remote installation costs range from 50 to 100 yuan per session.

In response, Zhang Siyuan, a special researcher at Su Commercial Bank, pointed out that the costs and installation prices for OpenClaw are relatively high, mainly due to its complex technical architecture and special requirements for computing resources.

First, as an AI agent framework, OpenClaw’s operation heavily depends on large language models (LLMs) as its “brain,” which requires continuous calls to cloud APIs or local deployment of high-performance models, both involving significant computing costs. Especially when handling complex tasks or high-frequency calls, API fees can quickly accumulate.

Second, Zhang Siyuan said that configuring a local deployment environment for OpenClaw is technically complex, involving multiple component integrations, permission settings, and network configurations. On-site paid installation services on social platforms are aimed at helping ordinary users with environment setup, dependency installation, and troubleshooting, which increases labor costs.

Furthermore, to ensure stable operation and responsiveness, users often need to configure high-performance hardware (like GPUs) and stable network environments, further increasing initial investment and maintenance costs. “Overall, the technical complexity, computing power demands, and professional deployment services together create a high cost barrier for ‘raising lobsters,’” Zhang Siyuan explained.

Market Cooling Down

While “raising lobsters” can greatly improve efficiency, security issues are also emerging. It is reported that “lobsters” have high system privileges, and improper configuration or attacks could have serious consequences.

A widely circulated case involved Summer Yue, a researcher from Meta’s superintelligence team, who shared a frightening experience on X (formerly Twitter): her “lobster” suddenly started deleting emails in bulk, and she could hardly stop it.

In response, the Cybersecurity Threat and Vulnerability Information Sharing Platform of the Ministry of Industry and Information Technology and the National Internet Emergency Center issued risk alerts regarding the safe use of OpenClaw, warning netizens to pay attention to information security.

Tian Lihui, director of the Financial Development Research Institute at Nankai University, told reporters that the security risks of OpenClaw stem from its disruption of the traditional “three-legged” security model. In traditional architectures, users, operating systems, and applications have clear permission boundaries, but OpenClaw, as a superintelligent agent, acts as a “master key” capable of opening all doors.

Tian Lihui explained that this over-concentration of permissions manifests in vulnerabilities like prompt injection and plugin poisoning at the technical level, and exposes a vacuum between open-source community autonomy and national security regulation at the institutional level. It turns cyberattacks from “data theft” into “controlling the physical world,” with a qualitative change in risk level.

The market has also responded. Reporters noted that currently, Wind (Wande) has not clearly classified “lobster” concept stocks, with market speculation mainly concentrated in cloud services and computing power sectors, but the hype has begun to decline. The most closely watched stocks include: UCloud-W (688158.SH), Qingyun Technology-U (688316.SH), Borui Data (688229.SH), Shunwang Technology (300113.SZ), and Hand Enterprise (300170.SZ).

Data from Wind shows that these five stocks have seen significant recent gains. From February 1 to March 10, excluding Hand Enterprise, the other four stocks rose over this period, with Borui Data leading at 65.78%.

However, on March 11, all five stocks experienced declines, with Borui Data dropping the most at 8.75% in a single day, and the others also falling to varying degrees.

Due to these issues, several listed companies recently issued risk warnings, reminding investors to watch for “unrealized performance” and “security risks.”

On March 10, UCloud issued an “Announcement on Abnormal Fluctuations in Stock Trading.” The company stated that its lightweight cloud host products based on OpenClaw images have not yet formed a scaled product system, and technological iteration and commercialization may not meet expectations.

UCloud also pointed out that frameworks like OpenClaw are still in early development stages, with uncertainties regarding market potential, technological stability, and data security. Multiple cloud service providers have launched similar products, intensifying market competition, and the contribution of related products to the company’s future performance remains highly uncertain.

Regarding the recent decline in “lobster” concept stocks, Zhang Siyuan told reporters that the correction on March 11 was mainly due to an emergency safety risk warning issued by relevant authorities. Monitoring found that some instances of OpenClaw open-source AI agents, under default or improper configurations, pose high security risks, easily leading to cyberattacks and information leaks.

Zhang Siyuan said that official risk alerts directly affected market sentiment, causing investors to worry about the short-term prospects of related stocks. OpenClaw’s features of autonomous decision-making and system resource calls, combined with fuzzy trust boundaries and a lack of strict review for skill packages, pose significant security hazards. “Such policy risk warnings often exert short-term suppressive effects on emerging technology sectors, especially when involving cybersecurity and data privacy concerns.”

In addition, Bi Mengran believes that the previous overheating speculation and profit-taking by funds are also reasons for the “lobster” sector’s correction. She pointed out that the “lobster” concept initially surged due to the “全民养虾” (nationwide lobster-raising) craze and token monetization expectations, with related sectors like computing power leasing and cloud services hitting daily limits. The rapid rise led to overvaluation. “With the security warnings, profit-taking funds took the opportunity to exit, causing a collective sell-off and a sector correction.”

Moreover, Bi Mengran noted that the commercial realization of “lobster” companies has fallen short of expectations, putting pressure on investment logic. Although the “lobster” craze remains high, most listed companies have disclosed that their related businesses have not seen significant growth so far. Investors have realized that earlier expectations were overly optimistic, lacking solid profitability, leading to profit-taking and sector decline.

“Raising Lobsters” Should Follow the “Sandbox Principle”

Due to numerous security issues, interviewees all advised against blindly deploying or using “lobsters” by ordinary users, as the technology and security systems are still under development.

Tian Lihui pointed out that this technology is still in the “laboratory stage,” not yet meeting the safety standards of consumer-grade products. If research purposes require experimentation, the “sandbox principle” should be followed: use completely isolated dedicated devices, configure credentials with the principle of least privilege, and strictly audit AI behavior logs. Before regulatory frameworks and technical standards are fully established, the best approach is to “observe” rather than “enter.” Respect for unknown forces may be more important than the courage to explore.

Angel investor and senior AI expert Guo Tao also told reporters that if netizens want to try “raising lobsters,” they should focus on: permission control, building isolated environments with sandbox and container technologies; command auditing, establishing input whitelist mechanisms to prevent prompt injection attacks; plugin governance, only using function modules from official channels to avoid malicious code; vulnerability response, tracking CVE (Common Vulnerabilities and Exposures) information in real-time, updating security patches promptly to reduce exploitation risks. For enterprise users, establishing security and compliance systems, clarifying usage scenarios and responsibilities is also necessary.

“From a broader perspective, the rapid rise of OpenClaw reflects the major trend of AI from ‘dialogue’ to ‘execution,’ but its security challenges also highlight the importance of AI governance. As technology matures and regulatory frameworks improve, AI agents are expected to play a greater role in a safer environment. But at this stage, users need to balance technological innovation with risk management,” Zhang Siyuan said.

From an investment perspective, Bi Mengran also advised ordinary investors not to blindly follow the “lobster” concept stocks. She stated that currently, “lobster” is still in the application exploration stage of AI iteration, with prominent security risks and less-than-expected commercialization. The short-term hype is mainly driven by capital speculation, and long-term investment value remains to be seen.