"Everyone raises shrimp", the awakening of big manufacturers
Large tech companies have finally realized that AI has truly transitioned from a "toy" for the elite to a "tool" for the general public.
In the early spring of 2026, a red "crayfish" crawled onto the screens of global developers and began to enter the lives of countless ordinary people.
The "crayfish" is called OpenClaw, formerly known as Clawdbot and Moltbot. In just a few months, it achieved the miracle of receiving 145,000 stars on GitHub. Shortly after the Spring Festival, various offline sharing sessions and online live broadcasts were successively held in Beijing, Shanghai, and Shenzhen, gathering people interested in OpenClaw. These people were afraid of missing any opportunities for learning, communication, and hands - on installation of the "crayfish".
In this "national crayfish - raising" craze, the personal experience of Fu Sheng, the CEO of Cheetah Mobile, became the most representative practical case. During the Spring Festival, Fu Sheng accidentally got injured while skiing, with a dislocated hip and had to stay in bed for recuperation. Unable to move freely, he spent only 14 days building a dedicated team of 8 AI Agents through remote voice control.
This AI team he called "Sanwan" runs automatically 24/7 without rest, efficiently and autonomously executes various tasks, and ultimately helped a WeChat official account produce a blockbuster article with over 100,000 views, and a Twitter post gain over one million views. It also simultaneously managed live - streaming and short - video operations, achieving over one million views.
The craze continues. On March 6, 2026, at an offline public - welfare installation event for OpenClaw at the north square of Tencent Building in Shenzhen, there was an influx of people who flew in from all over the country to learn about the deployment.
To date, this movement known as the "national crayfish - raising" is reshaping our cognitive boundaries of artificial intelligence at an unprecedented speed. However, while the technology is iterating, we also question whether this is just an experiment with a huge bubble. Some voices claim that there is a sense of "fleece - cutting" behind this national frenzy.
It's true that some people without technical knowledge blindly follow the trend, and many application scenarios are still immature, even accompanied by security risks. But if we look beyond the noisy surface, we'll find that this is not just a technical or capital carnival; it's also a "national infrastructure project" for AI socialization.
In this national movement, the roles of large tech companies are also shifting. They have finally realized that AI is truly transitioning from a "toy" for the elite to a "tool" for the general public.
01 From "Elite Toy" to "National Experiment"
In the years since entering the AI era, in the perception of many people, large - scale models have been the exclusive domain of a few tech geeks and research institutions. However, with the explosion of OpenClaw, this cognitive barrier is being broken.
According to public reports, OpenClaw is an open - source AI agent mainly designed to help users handle daily tasks such as sending emails and managing calendars. It has received over 145,000 stars on GitHub and is regarded by many business leaders as the "future engine" capable of independently running an entire company.
What's even more astonishing is its spread speed. The project was officially launched at the end of 2025. By the end of January 2026, Tencent Cloud and Alibaba Cloud had successively launched one - click cloud deployment solutions, completing ecological adaptation and on - the - ground support. This craze quickly spilled over from the tech circle and even triggered a kind of almost pilgrimage - like fanaticism among ordinary people.
A relevant person in charge of Tencent Cloud revealed to "Business Show" that on March 6, he witnessed a shocking scene on - site. Many developers, entrepreneurs from all over the country, even some corporate employees, especially those from Hangzhou, flew to Shenzhen one day in advance just to experience and deploy OpenClaw as soon as possible.
These "pilgrims" who came from across cities were not here for a closed - door tech summit. They just wanted to get hands - on guidance at this deployment site, and also seemed eager to bring this new intelligent agent architecture back to their teams or help their own scenarios better implement. "I want it to help me trade stocks better," said a person on - site.
This "pilgrimage" scene was concretely presented in the offline event, and this detail also indicates that the power structure of technology is undergoing a fundamental reversal. "Actually, I think many people might not even know what the crayfish is, but they still came directly," the aforementioned interviewee told "Business Show".
From the live - stream footage of the Shenzhen event, we can see that hundreds of developers arrived early with their laptops and queued up in the square. The queue snaked into three columns, almost taking up a corner of the entire square. It is reported that there were not only local geeks on - site but also a large number of users who came specifically from other places.
To cope with the bustling scene, the Tencent Cloud team set up nearly 10 long tables as temporary deployment areas. Next to each table, there were 3 technical experts sitting face - to - face with users, providing hands - on, nanny - level teaching from local environment configuration to Docker container startup.
This atmosphere of "going to battle with weapons" shows the users' urgent need to eliminate deployment barriers and obtain a deterministic environment. We even saw a developer working in a nearby science park asking in the live - stream group, "What time does the event end? Is it okay to come during lunch break?" It seems that this corporate employee might only come to complete the deployment within just one or two hours and then go back to work with this "crayfish" in the afternoon.
This time - conscious mindset confirms the working people's eagerness for AI to improve efficiency. When someone said in the group that the on - site queue was very long and everyone should deploy according to the tutorial by themselves, some people said they were not willing to wait for the online tutorial in the group and fumble slowly. Instead, they hoped to quickly complete the environment configuration with the help of on - site experts during fragmented time and immediately connect OpenClaw to office scenarios such as Feishu and WeCom.
Public reports show that since late January 2026, the download volume from Chinese IP addresses has increased sharply, and the average daily page views of the project's Chinese documentation have exceeded those of all other non - English languages.
According to the recent observation of "Business Show", even more, a large number of non - technical people have come from other places to first - tier cities like Beijing and Shenzhen just to experience the fun of "raising crayfish". They might not understand what the Transformer architecture is, nor care about the algorithm logic of Token consumption, but they come with the most basic needs, as long as it can help them automatically organize emails, monitor inventory, or even just have a chat to relieve boredom.
We found that there are also many content creators in this group. They hope to hand over everything from content creation to operation, release, and even later - stage data monitoring and integration to their super AI individual assistants. For most people, it doesn't matter whether it can be achieved or how well it can be done. The first thing is to deploy the "crayfish".
This is also the focus of the controversy over the "bubble". Some people think that the influx of a large number of people without a technical background might bring chaos and inefficiency. However, the product logic behind the explosion of OpenClaw lies in that it truly realizes "messages as instructions", naturally combining the chat experience of ChatBot with an agent layer that can be triggered, has permissions, and can execute actions. In addition to the founder's product concept, another topic that has attracted industry attention is that OpenClaw has also brought benefits to both hardware and software directions, and even caused a short - term supply shortage of Mac Mini due to deployment requirements.
The imagination of the elite in large tech companies has limits, but the needs of the general public are infinite. When AI moves from "conversation" to "execution", only a large number of real - world scenarios can train true intelligence. Those seemingly absurd "wild" usages are actually exploring the unknown boundaries for AI agents.
02 Security Alarm: Putting an "Electronic Ankle Bracelet" on AI
There's no denying that there is a bubble in the current OpenClaw craze, and there are also huge potential security risks hidden behind it. This is also the core reason why many large tech companies are vigorously promoting cloud - based deployment rather than local deployment.
We know that OpenClaw is different from traditional chatbots. It has extremely high system permissions. It can read and write files, execute Shell commands, control browsers, and even operate email accounts. Once it is deployed locally on a personal computer, it's like giving the keys to your home to an immature "digital employee".
Moreover, there have been several astonishing security incidents recently. After some users deployed OpenClaw locally, due to prompt injection or configuration errors, OpenClaw accidentally deleted important local documents or even formatted the corresponding folders.
Some cybersecurity experts have warned that if a malicious attacker hijacks the local OpenClaw through hidden instructions embedded in web pages or emails, the consequences could be disastrous. At best, it could lead to privacy leakage; at worst, the entire personal computer could become a "zombie computer".
Therefore, "keeping the Agent in a cloud - based cage" has become an industry consensus. The lightweight application server solutions launched by Tencent Cloud and Alibaba Cloud are essentially building an isolated operating environment for this high - permission AI agent.
For example, through physical isolation, the Agent runs in a cloud - based virtual machine, completely physically separated from the user's personal computer and mobile phone. Even if the Agent is compromised, goes out of control, or makes an accidental operation, the only loss is a cloud - based instance that can be reset at any time, and it will never affect the user's local photos, documents, and bank accounts.
And the permissions are controllable. The cloud - based environment can strictly limit the Agent's network access and file access scope, following the principle of "sensitive information should not be on the Internet, and information on the Internet should not be sensitive". There is also the ability to roll back at any time. Once there is a configuration error or an attack, the cloud - based image supports second - level snapshot recovery, while local data is often difficult to recover once lost.
A technical expert from a large tech company described it to "Business Show" like this: "We encourage everyone to 'raise crayfish', but you mustn't let the crayfish 'run' around at home. Cloud - based deployment is like a transparent 'fish tank' that allows you to see how it works while ensuring that it won't jump out and knock over your 'furniture'."
03 From "Money - Burning" to "Inclusiveness"
Besides security, cost is another practical barrier. OpenClaw needs to frequently call large - scale models when performing tasks, consuming a huge amount of Tokens. It can be called a "Token burner".
Fu Sheng, the CEO of Cheetah Mobile, shared his actual measurement data in a recent live - stream: As a heavy user, he spends about over $100 on OpenClaw every day because his "crayfish" team undertakes high - intensity commercial tasks such as WeChat official account operation and live - streaming data review. However, Fu Sheng also pointed out that for ordinary individual users, if they don't use it in a heavy - duty commercial way, a monthly expense of dozens of dollars is enough for daily use.
More importantly, the cost of infrastructure is rapidly decreasing, further reducing the usage threshold. According to the official event information of Tencent Cloud, the customized package of lightweight application servers for developers is priced as low as $7.9 per month. This two - way cost optimization of "online and offline", combined with the security guarantee provided by the cloud, has turned "raising crayfish" from a game for the rich into an ordinary person's daily activity.
However, bubbles are often a necessary "fermentation" process before the popularization of new technologies.
Firstly, bubbles bring about rapid expansion in scale. Data shows that the global AI Agent market size will soar from $5.1 billion in 2024 to $80 billion in 2027. The annual compound growth rate of the Chinese market is expected to exceed 60%, reaching 171.65 billion yuan by 2031. Without this explosive attention and participation, the education cost of AI agents would be prohibitively high.
Secondly, bubbles also accelerate the screening of application scenarios. Peter Steinberger, the founder of OpenClaw, said bluntly in an interview that he couldn't stop using OpenClaw and felt that he no longer needed any independent apps. From the way he saw people using OpenClaw, there was a sign that most apps would disappear. "Why do I still need a fitness app?" he said. "I just need to send it a photo of my food, and it can match all the information, adapt to my routine, and ensure that I can achieve my goals."
This path of "execute first, optimize later, and then reconstruct" is the process of screening out the 10% of "Financial Agents", "HR Agents", or "Research Partners" customized based on real - world pain points among millions of attempts.
An industry insider told "Business Show", "Actually, the more people know about it, the higher the possibility of finding effective application scenarios. " Obviously, this "national trial - and - error" mechanism is actually a way to complete a stress test and scenario verification of AI agents for the whole society at the lowest cost.
04 The Role Transformation of Large Tech Companies: From "Creator" to "Shovel Seller"
In this national movement, the roles and perceptions most affected are those of large tech companies. In the past, large tech companies were accustomed to playing the role of "creator", trying to build all - powerful super - models and restrict users within their own ecological walls.
However, the rise of OpenClaw has forced them to re - position themselves and make strategic deployments.
MiniMax, a mainstream general artificial intelligence company in China, has built a diversified product matrix. On January 27, 2026, MiniMax officially announced that it had provided technical support for OpenClaw in terms of model capabilities. Combining with the actual implementation cases of overseas developers, OpenClaw with MiniMax M2.1 as its core engine performs excellently in tool - calling accuracy, task - execution efficiency, and cost control.
In terms of model - calling requirements, most users have reported that "it burns too many Tokens". The direct impact of this is that compared with the most expensive flagship models of model providers such as OpenAI, Anthropic, and Gemini, it is more suitable to use the cost - effective model APIs for simple tasks on OpenClaw. Among them, Minimax and Kimi, two Chinese providers with relatively good overseas services, have been recognized.
Facing this trend, large tech companies have different responses but end up with the same result. First, OpenAI chose to "recruit". On February 15, OpenAI announced that Peter Steinberger, the founder of OpenClaw, had joined the company to be responsible for the development of the next - generation personal Agent. The OpenClaw project will be transferred to an independent foundation for operation and will receive resource support from OpenAI.
Domestic providers have also transformed into "shovel sellers". On January 28, 2026, Tencent Cloud and Alibaba Cloud successively launched exclusive cloud - based deployment solutions for Moltbot (the former name of OpenClaw), supporting users to achieve one - click installation and rapid deployment through lightweight servers. They are no longer obsessed with controlling terminal applications but instead provide underlying model computing power and convenient access tools, willing to be the infrastructure.
Behind this role transformation is also the continuous renewal of the large tech companies' perceptions. As the aforementioned person from Tencent Cloud sighed to "Business Show", "We didn't realize it before. We used to think it was a toy for the elite, but now we find that it's something for the whole nation. We must remove this threshold as much as possible to let more people in."
Obviously, they have realized that in the Agent era, the real moat is no longer the model itself but the huge application ecosystem and massive data feedback built on the model. Only when more scenarios and data are connected can there be a chance to develop amazing products and solutions.
OpenClaw started to soar in February this year, with more than a dozen updates in 20 days, and it connected to Gemini 3.1 and Apple Watch, showing strong growth momentum. Data shows that within just two weeks, the Token usage of OpenClaw in the open - source weighted model field has accounted for 13% of the total on the OpenRouter platform. This open attitude has quickly made it a connector for the model ecosystem and also made large tech companies realize that "connection" has greater value than "exclusivity".
Conclusion
The "crayfish storm" brought by OpenClaw may eventually subside, and that red icon may no longer be as vivid. But the era driven by intelligent agents that it has initiated has just begun.
The development path revealed by OpenClaw contrasts with the current development focus of some domestic AI applications. The former is obviously more specific, aiming to solve systematic efficiency pain points such as cross - platform information integration and automated processes, while the latter, as we've seen during the