Raising 'lobsters' drives up computing power demand! Where are the investment opportunities?

贝克街探案官

joined discussion ·

A 'lobster' ignites the 'ancient AI battleground,' potentially creating a new employment boom

Baker Street Investigator

As agents begin to take over human tasks, token consumption is entering an exponential era

Author：Bao Kemo

In early 2026, a 'red lobster' suddenly went viral in global tech circles and on Chinese social media platforms. It wasn't a new type of pet or internet meme but rather a familiar 'new star in AI agent tools'—OpenClaw, due to its red lobster icon, domestic netizens quickly joked about deploying and using it as 'raising lobsters'.

This originally technical project within programmer communities rapidly spread from GitHub tech circles to government affairs, enterprise offices, and even everyday discussions among regular internet users within just a few months, becoming one of the most talked-about AI phenomena of 2026.

Shortly after OpenClaw was launched, its number of stars on GitHub skyrocketed almost linearly in an extremely short time. As of March 8, OpenClaw had reached 260,000 stars and nearly 48,000 forks on GitHub, surpassing projects like React and Linux, with its popularity continuing to rise.

Its explosive popularity is also reflected in third-party model API routing platform data,During the period from February 5, 2026, to March 5, 2026, OpenClaw was the application with the highest token consumption on the OpenRouter platform,with its token consumption reaching as high as 7.63 trillion, far surpassing the second place.

In addition to OpenClaw, major model developers have also increased their investments in Agents, with AI Agents now entering the practical implementation phase. For example, Minimax Agent can not only handle files like PPT, Excel, Word, and PDF but has also launched desktop and expert Agents. The desktop version allows the Agent to enter the user's work environment to directly help organize files and streamline information. Expert Agents, after being injected with specific knowledge and behavior templates and learning detailed SOPs, can perform more effectively in certain types of tasks.

The explosive popularity of OpenClaw is just the tip of the iceberg for the industry embracing intelligent agents.Behind it lies a transformation in the industry and an explosive growth in supporting infrastructure.。

01 A 'lobster' drawing widespread attention from all sides

The core capability of OpenClaw is transforming AI from merely a chatting tool into a truly autonomous 'task-performing' intelligent agent.

The rapid rise of OpenClaw is closely tied to its open-source nature. The project was developed by Austrian software engineer Peter Steinberger, with early versions named Clawdbot and Moltbot, which gradually evolved into OpenClaw.

While traditional AI is merely a 'talking brain,' OpenClaw is more akin to a 'thinking and hands-on digital assistant.' Users simply need to specify goals, such as 'organize this week's customer complaints and generate an analysis report,' and the system will automatically open spreadsheets, organize data, invoke models for analysis, and generate documents, achieving cross-software automation.

After users deploy it locally and connect it to a large language model, OpenClaw can take control of the user's computer mouse and keyboard operations, breaking through barriers between various software and platforms to execute tasks. These include automatically organizing documents, batch processing data, replying to emails, generating reports, and even automating complex workflows across multiple applications.

Because of these capabilities, OpenClaw has quickly found real-world application scenarios across various industries.For instance, in the public sector, some civil servants in Shenzhen have begun testing a version called 'Government Lobster' to handle citizen requests and administrative licensing guidance. By automatically organizing citizen complaints, categorizing issues, and generating handling suggestions, efficiency has been significantly improved. In corporate office environments, many teams use it for report organization, customer service replies, and operational data analysis, greatly reducing repetitive tasks.

Following Longgang District in Shenzhen and the High-Tech Zone in Wuxi, Changshu City in Suzhou has also built a 'Crayfish Pool' to attract people to 'raise crayfish' locally. According to an official WeChat post by the city government's news office 'iChangshu' on March 9, the city has released the 'Several Measures (Draft for Comments)' aimed at accelerating the development of open-source communities such as OpenClaw to promote high-quality industrial growth. The measures include 13 initiatives to support OPCs (one-person companies) in using OpenClaw for business operations, offering up to 6 million yuan in comprehensive support for OPC projects selected under various talent programs.

At the recently held national conference, several delegates and committee members also mentioned this phenomenon. Zhou Hongyi, founder of 360 Group, stated that intelligent agents similar to OpenClaw are transforming 'cloud-based software capabilities into everyone’s personal assistant.' The explosive popularity of this technology even exceeded Ma Huateng’s expectations.

In March 2026, Tencent provided free OpenClaw installation services at its Shenzhen headquarters, attracting thousands of people to queue for the experience. During the same period, Longgang District in Shenzhen also issued specific policies to support the OpenClaw ecosystem, offering up to 1 million yuan in rewards for related innovative application projects, aiming to convert this open-source project into new industrial opportunities. From the government, enterprises to individual developers, a completely new technological ecosystem is forming around OpenClaw.

On the other side of this technological boom, a gray market service revolving around 'raising crayfish' has quickly emerged. Since OpenClaw requires local deployment environments, model configurations, and permission settings, it presents a high barrier for ordinary users, prompting many technicians to offer installation services.

On domestic social media platforms, various‘door-to-door crayfish raising’ and ‘remote crayfish raising’ads quickly appeared, with prices ranging from tens to hundreds of dollars: remote installation services typically cost between 50 to 100 yuan, while door-to-door setup and debugging range from 300 to 1,000 yuan. Some practitioners even claimed that by helping others install OpenClaw, they earned an income of 260,000 yuan within days.

But for individual users, this installation fee is just the beginning; subsequent server rental charges apply, starting at 50 yuan per month, with about 200 yuan needed for a smoother configuration. Building your own 'crayfish trap' could easily cost 4,000 yuan for a decent setup. However, hardware alone isn’t enough to keep your 'crayfish' ready to go at all times—you must continuously provide tokens to support the crayfish’s operation. Entry-level costs start at 30 yuan, with no upper limit. The smarter the model you use, the higher the expenses, and having your crayfish complete basic tasks could cost upwards of 10,000 yuan monthly. In response, some netizens remarked: 'As long as your traps are novel enough, there will be endless韭菜to harvest.'Some netizens commented: 'For ordinary people, it’s completely useless.'

Source: Silicon Intelligence Agent video account

In addition to the high costs, for individual users,raising lobsters also poses significant security risks.。

02 Security Concerns of Raising Lobsters

OpenClaw has the ability to run continuously, make autonomous decisions, and call system resources. If improperly configured, its risks far exceed those of ordinary software.

The Cybersecurity Threat and Vulnerability Information Sharing Platform of the Ministry of Industry and Information Technology recently found during monitoring that,some OpenClaw instances posed a higher security risk under default or improper configurations.When deploying OpenClaw, the trust boundary is often unclear, and the system typically has high-level permissions. Once malicious instructions are induced or taken over by attackers, it may perform unauthorized actions, leading to sensitive information leakage, remote system control, or even becoming a node for cyberattacks.

Firstly, there is the risk of permission abuse, where AI might access core corporate data or delete critical files under instruction manipulation; secondly, the risk of information leakage, as open-source versions lack robust data encryption and access auditing mechanisms; thirdly, the risk of malicious takeover, where some instances exposed to public networks could be exploited by hackers, serving as an entry point to attack enterprise systems; fourthly, the risk of unpredictable behavior, such as accidental data deletion, automatic execution of unauthorized operations, or even being used by criminals for spam distribution and online fraud.

Therefore,Experts recommend deploying OpenClaw in isolated environments or cloud-based sandboxes,disabling unnecessary public network access, and establishing identity authentication, permission management, data encryption, and security auditing mechanisms to reduce potential cybersecurity risks.

The security risks highlight that the biggest difference between AI agents and traditional software lies in the fact that AI agents can not only be 'used,' but also possess a certain degree of 'autonomous action capability.' Without robust access control, identity authentication, logging, and data encryption mechanisms, the exploitation of such systems could lead to consequences far more severe than vulnerabilities in ordinary software.

Therefore, authorities recommend that organizations and individual users deploying OpenClaw disable unnecessary public network access interfaces, strictly configure identity verification and access control mechanisms, encrypt data, establish a security auditing system, and continuously monitor official security announcements and reinforcement recommendations to mitigate potential cybersecurity risks.

However, from an industry perspective, any technological innovation is a double-edged sword—where there are risks, there are also advantages. The 'lobster farming' phenomenon reflects a transformation in the form of artificial intelligence applications.AI is transitioning from being a 'conversational tool' to becoming an 'action agent.'In the past, people primarily used AI to ask questions and receive answers, but now AI has started directly participating in workflows and executing specific tasks.

When an intelligent agent capable of understanding objectives, planning steps, and autonomously operating software emerges, the relationship between individuals and computers is also changing. As many industry insiders have said, the explosive popularity of OpenClaw isn't just the success of an open-source project—it may very well be a prelude to a transformation in working methods in the AI era.

If the keyword of the internet era was 'connection,' and the keyword of the mobile internet era was 'platform,' then in the age of AI agents, a new keyword is emerging—'agency.' As more and more tasks are completed by intelligent agents, the way humans collaborate with technology will be redefined.

And that accidentally viral 'red lobster' may just be a signal marking the beginning of this new era of AI agents. The tokens that ordinary people cannot afford are precisely where AI companies will see future revenue growth.

Section 03: Industry Transformation Brought by 'Lobster Farming'

As agents become more prevalent in office work, development, operations, and other scenarios, a large number of tasks will be completed through AI automatically invoking models, tools, and data interfaces.This indicates that computational power and token consumption will continue to grow.。

According to IDC forecasts, the number of active agents in China will exceed 350 million by 2031, with an annual compound growth rate of over 135%, leading global growth. As task density and complexity increase, agent token consumption is expected to experience exponential growth, rising more than 30-fold annually.

Currently, token consumption on the enterprise side in China's market remains dominated by conversational and generative AI. However, as the scale of agent operations and task complexity rise simultaneously, token consumption for active agents is entering a period of rapid growth. According to the First Sounding Think Tank, based on enterprise agents' token consumption, breadth of scenario coverage, and impact on core business, agents can be divided into five levels. China is currently transitioning from widespread adoption to integration, with annual token usage still having a potential increase of 10 to 100 times. Chinese agents are on the verge of explosive growth with significant room for expansion.

From personal assistants to corporate 'digital employees,' every task execution, data analysis, and process collaboration by AI agents requires invoking model capabilities for support. As the number of agents and frequency of use continue to rise, token consumption is likely to experience exponential growth, driving the formation of a new AI application ecosystem.

As one of the leading large-model vendors in China, ByteDance has delivered impressive results in the AI field as the year comes to a close. By December 2025,the daily average token usage of DouBao's large model has exceeded 50 trillion, increasing more than tenfold compared to the same period last year. Currently, over 100 enterprise clients have cumulative token usage exceeding one trillion.

According to research by EqualOcean, the penetration rates of AI agents in key account (KA) and small and medium-sized businesses (SMB) will rise from 3%/0.5% in 2023 to 25%/10% by 2028. The market size of China’s AI agents will also expand rapidly at a compound annual growth rate of 125%, surging from 57.4 billion yuan in 2023 to 3,300.9 billion yuan by 2028.

According to a speech by Yang Chaobin, CEO of Huawei ICT BG, global daily token consumption has grown nearly 300-fold over the past two years. According to the STAR Market Daily, China’s overall daily token consumption was 100 billion at the beginning of 2024, surpassing 30 trillion by mid-2025. By February 2026, the combined daily token consumption of mainstream large models reached 180 trillion tokens.

Refocusing on OpenClaw, the token consumption published on the Open Router platform shows that token usage quadrupled within a month. Notably, it has been less than two months since OpenClaw's official launch on January 29, 2026.

According to user data from the Open Router platform, the token consumption of the OpenClaw application surged from 80.6 billion on February 3, 2026, to 358 billion on March 4, 2026. Its token consumption increased approximately 4.4 times in just one month, highlighting the massive demand AI agents have for tokens.

Due to the inherent characteristics of multi-tool invocation, long context, and multi-process workflows, the growth rate of token consumption by AI agents is extremely rapid. According to research estimates, compared to basic word dialogue bots, the computational power demands for image generation, reasoning, video generation, and deep research are 10x, 100x, 3000x, and 1,000,000x respectively. As task complexity increases, computational power requirements grow exponentially, driving a significant leap in token consumption.

AI agents perform tasks by invoking various tools. During operation, a common paradigm is to break down instructions into different processes and handle them in stages, which inherently involves long contexts. This paradigm leads to a very rapid increase in token consumption for AI agents.

Conclusion

The core expenditure of intelligent agents is concentrated in the reasoning phase. As agent penetration increases, the focus of computational power demand shifts from training to reasoning. Reasoning computational power primarily handles inference tasks for AI models, focusing on executing pre-trained models with an emphasis on low latency and low power consumption.

As AI transitions from being training-focused to reasoning-focused, there is a surge in demand for private environment and edge deployments. According to IDC data, as AI moves towards prioritizing reasoning, the demand for reasoning scenarios continues to rise, with the share of reasoning workloads expected to increase from 65% in 2024 to 73% in 2028.

With the expansion of AI reasoning needs, China’s reasoning market is projected to grow to 293.12 billion yuan by 2028. AI applications are shifting from a lack of model training to a shortage of reasoning services. Reasoning computational power, as a core infrastructure for AI commercialization, will directly benefit from the industrialization of intelligent agents and the deployment of large industry-specific models.

According to Frost & Sullivan and LeadLeo Research Institute, the scale of China’s reasoning market was 17.52 billion yuan in 2024 and is expected to grow to 293.12 billion yuan by 2028, with a compound annual growth rate of approximately 102% from 2024 to 2028.

Against this backdrop,Electricity demand and computational power requirements are bound to see explosive growth,which warrants continued attention from the market.

All materials sourced from official public information

This article does not constitute any investment advice.

11K Views