OpenClaw Carnival, Capital Calmness

If your WeChat Moments weren't flooded with posts about "installing lobsters" this March, you might not be in the tech circle.

A reporter recently signed up for an offline "free lobster installation" event. The venue, originally planned for 300 people, quickly had over 400 sign - ups, and the organizers had to close the registration channel ahead of schedule. From geek communities to public platforms like Xiaohongshu and Douyin, installation tutorials and user experiences are everywhere. Big tech companies are getting involved intensively - Tencent's WorkBuddy, ByteDance's ArkClaw, Alibaba's CoPaw, and Xiaomi's miclaw, with more and more fancy names.

OpenClaw, an open - source AI Agent jokingly called the "crayfish", is going viral at an amazing speed. It has evolved AI from "just talking" to "getting things done" - directly operating the computer, invoking tools, and completing actual tasks through natural language instructions. An investor described it like this: "It turns tools into workers. I work 8 hours a day, while it works 24 hours a day."

While geeks are having a blast over the "digital employees", the venture capital circle, which is closest to money, is relatively quiet.

Everyone is installing, everyone is waiting

Almost all the investors interviewed by the reporter said that they had installed OpenClaw and were in the process of experiencing it.

Young VC investors are installing it, the heads of state - owned institutions are installing it, and FA institutions are even asking their technical teams to try to optimize the configuration. Some people are posting screenshots of successful installations on their WeChat Moments, while others are complaining that "it hasn't worked after two days of installation".

But what after the installation? Everyone is waiting and watching.

Different FA institutions gave similar feedback: "Members of our team are all trying to use it, but no relevant projects have entered the trading scope." Early - stage VCs are also cautious - some are using it, observing it, and discussing it, but "it's not the time for investment yet". State - owned institutions are even choosing to stay on the sidelines. Only one dual - currency fund clearly said that their team is paying attention to such projects.

Everyone is installing, but it seems that everyone is waiting.

This wait - and - see attitude mainly comes from security concerns. More than one investor expressed their concerns to the reporter. OpenClaw has extremely high operating permissions - it can perform any operation, access local files, invoke system commands, and connect to external APIs with authorization. A VC investor who is debugging it admitted: "I don't even dare to let it run freely on my own computer, let alone recommend it to the invested companies."

The Ministry of Industry and Information Technology has previously issued a security warning for OpenClaw, pointing out the risk of its "fuzzy trust boundary". Exposed control interfaces, abuse of execution - layer permissions, and supply - chain security of the plug - in ecosystem - these issues may be tolerable in the geek circle, but in a business scenario, any one of them is a fatal flaw.

"It's okay for fun, but what if the company loses data if it uses it?" an investor asked rhetorically.

Another concern comes from stability. A VC partner who is trying the cloud - based version complained to the reporter: "It often reports errors and overloads. Whenever I send a task to the AI, it often tells me that the system is down. How can such an experience be commercialized?"

This state of "everyone is installing, but everyone is waiting" exactly reflects the essence of this wave of enthusiasm: the technology has indeed brought breakthroughs, but there is still a long way to go before a real business closed - loop is formed.

To understand this gap, we need to first understand what OpenClaw has actually done. In essence, it is an AI Agent execution framework that prioritizes local use and is open - source and free, evolving AI from "dialogue" to "execution". Liu Jin, a partner at Yida Capital, analyzed for the reporter: "Previously, large models remained in the dialog - box mode, not much different from search engines. But OpenClaw has achieved the functional implementation of the intelligent agent, which is the real start of the implementation of large models in the vertical business field."

This explains why it has become so popular. It touches on the most basic desire of ordinary people: to let AI really help them with their work. Moreover, it has lowered the threshold for entrepreneurship from "writing code" to "being able to talk" - you don't need to understand APIs or code to develop your own applications.

But what the capital sees is another story. "OpenClaw is an open - source project on GitHub and doesn't have the concept of charging in itself," he admitted. "Similar to Linux, it's difficult to form a large market as a business. It's hard to form a business closed - loop on the To C side."

The commercialization dilemma of open - source projects is the first reason for investors to wait and see. Users are willing to pay for "results" but not for "frameworks". What big tech companies are doing is to cloud - enable OpenClaw. Alibaba Cloud and Tencent Cloud have launched their own "lobster" versions, which in essence are still selling cloud services. Start - up companies are caught in the middle and can't find an anchor point for charging.

"Chinese cloud providers are too strong. Tencent and Alibaba can develop their own versions based on OpenClaw and then integrate them into the cloud ecosystem for free. How can start - up companies compete?" Liu Jin said bluntly. Big tech companies have enough capabilities for redevelopment and optimization, and their optimizations may not be open - sourced. For secondary - development teams that rely on open - source, this is a blow at a higher dimension.

Moreover, what the commercialization scenario needs is not just "usable" but "easy to use" - stability, security, computing - power scheduling, and fault rollback, each of which is a tough nut to crack. "It's easy to make it run, but it's hard to make it run well."

But the most fundamental reason is the unclear exit path.

Investors from different institutions all mentioned that the domestic capital market has limited support for pure software business models. Especially during the wave of enterprise - service investment from 2015 to 2020, many institutions are still unable to get out of the quagmire. This history has made investment institutions naturally vigilant about pure software business models.

"The core of investment is to exit," Liu Jin said. "There is no clear IPO or acquisition path for this business model at present."

The feedback from FA also confirms this: although the internal members and technical employees of the team are all trying to use OpenClaw and have noticed the popularity of this direction, "no projects have entered the trading scope". A person from an FA institution summed it up directly: There is popularity, but no projects.

This state of "there is popularity, but no projects" exactly shows a problem: OpenClaw does have user value, and people are willing to install it, use it, and discuss it. But whether it can generate projects with capital value is still unknown at present.

Who is making money? The "shovel sellers" behind the craze

However, while investors and FA institutions are calmly waiting and watching, another group of people have quietly started making money.

OpenClaw has a characteristic: it is a "Token consumption amplifier". A single complex task may consume hundreds of thousands or even millions of Tokens. This means that model providers are the first to benefit. Zhipu, MiniMax, and Yuezhianmian, these large - model companies benefit most directly. A heavy user complained to the reporter that the monthly Token cost is "at least over a hundred, more expensive than a movie and TV VIP".

Cloud providers are also making money quietly. The offline "free lobster installation" event that the reporter participated in, which was overcrowded, was hosted by a cloud provider. On the surface, the event is a technical sharing, but the real purpose is to acquire customers accurately - paving the way for subsequent paid services. The service list provided by the host includes: optimizing GEO keywords on major AI platforms, using intelligent agents for market promotion, and making short videos three times faster with AI systems... After a series of combined operations, the goal is to draw customers into their own ecosystem.

There are also more direct "shovel sellers": on - site deployment service providers charge 300 to 1000 yuan for one installation; AI training courses teach people how to use OpenClaw to improve work efficiency.

This scene seems familiar. During the Linux wave back then, the ones who really made money were not Linux itself, but technology service providers like RedHat and the entire ecosystem that grew around it. OpenClaw is repeating this history - the most talked - about is the framework itself, but the ones who make the most money are always the surrounding "shovels".

So, do start - up companies have no chance at all? Not necessarily.

Based on the views of many investors interviewed, the opportunities are mainly concentrated in three directions.

The first direction is in - depth skills in vertical fields. For example, in professional scenarios such as financial auditing, medical documentation, and legal contract review, barriers can be established through industry know - how. "Big tech companies build general platforms but don't understand specific business. If you can penetrate a certain scenario, you will have room for survival," Liu Jin further analyzed. "Moreover, such scenarios often involve professional terms and complex processes, which cannot be solved simply by invoking APIs. It requires a real understanding of the business."

He gave an example: an Agent for financial auditing needs to understand accounting standards, auditing processes, and common financial fraud patterns. These are not naturally available in general large models and require high - quality data training and in - depth industry accumulation. "Whoever can accumulate this data will have a moat."

The second direction is private deployment in sensitive industries. In fields such as government, military, finance, and healthcare, where data must be localized, there are strict requirements for security and compliance. These customers are willing to pay for security and have sufficient budgets.

Some investors also mentioned that many institutions bought Deepseek all - in - one machines a long time ago but didn't know how to use them. OpenClaw can help them activate these computing resources and really put the models into use. In this field, whoever can help them complete this closed - loop will have value.

The third direction is engineering capabilities. Polishing the open - source framework to the "commercially available" level - stability, security, multi - cloud adaptation, computing - power scheduling, log management, and fault rollback. These dirty and tiring jobs may not be what big tech companies are willing to do, but B - side customers are willing to pay for them.

At the same time, many investors also reminded: The middleware layer may be very dangerous. Because AI can directly write code and directly invoke interfaces, the traditional 'tool layer' is being emptied. "In the past, middleware was used because of interface barriers and data - format barriers. Now AI doesn't care about these. It directly wants data and results. It's very difficult for this layer to become an independent business model.

The future start - up form may be a "one - person company" or a very small team, which can quickly test and monetize through unique data advantages and scene understanding. However, even if the business model is successful, the attractiveness of such companies to investment institutions is still limited. "They may not be large in scale but have good cash flow, but 'small and beautiful' doesn't equal 'good investment target'," an investor from an early - stage institution admitted. "We need to exit, and they need to grow. It takes time to bridge this gap. Moreover, the current preference of the capital market is for hard technology and heavy assets, and it's not easy to tell a story about pure software."

In his words, "Openclaw may have opened up a new business model, but from the perspective of the investment industry, whether it's a good investment target still needs to be observed. It won't be clear soon. Everything is just beginning."

This article is from the WeChat official account "Venture Capital Daily", author: Yu Shiqi. Republished by 36Kr with authorization.