To be honest, the growth rate of this project made me think the data was wrong when I first saw it.
It’s 8 degrees Celsius and cloudy in Shanghai today, perfect for staying indoors and mining data. I just came across a piece of news—nearly a thousand people lined up outside Tencent’s headquarters waiting to install the “Lobster” for free. An AI tool that can cause such a massive queue in the real world means something has indeed fundamentally changed.
Banned by Google and Anthropic, Chinese Companies Say: Thanks, We’ll Take Them All
Let me share a fact that many people might not know.
OpenClaw is the most explosive open-source project at the start of 2026. It garnered 250,000 stars on GitHub in just 4 months, surpassing Linux and React, becoming the fastest-growing project in human open-source history. What is it? Simply put, it’s an “AI worker living in your computer”—it can schedule your itinerary, write code, organize files, send emails, and operate browsers, working 24/7 without stopping. Moreover, all data is stored locally, with nothing uploaded to the cloud.
Its founder, Peter Steinberger, is an Austrian who previously spent 13 years building PDF tools. Thanks to this project, he was personally recruited into OpenAI by Sam Altman. Mark Zuckerberg also approached him, but they couldn’t reach a deal.
So, how did US tech giants react?
Google—internally banned, employees are not allowed to use it. Anthropic—also banned. Meta tried to poach him but failed. The top three US cloud providers? None have launched an official cloud service for OpenClaw.
What about China?
Alibaba Cloud, Tencent Cloud, Volcano Engine (under ByteDance), and Baidu AI Cloud—almost within the exact same time window, all launched one-click deployment services for OpenClaw. Baidu even ran a promotion for just one cent for the first month.
I chatted with a friend in cloud computing about this, and he nailed it: “Silicon Valley is still holding meetings to discuss whether to integrate it, while China is already competing on who can integrate it faster.”
Two Weeks, 30+ Products: How the Math Works Out
Let’s look at a number. From around the Spring Festival to late February, in about two weeks, more than 30 AI Agent products based on or benchmarking OpenClaw emerged in the Chinese market.
Over 30 products. In two weeks.
I dug into the main ones:
Alibaba Cloud CoPaw — Open-source, free, deployed with just three lines of code. It integrates seamlessly with DingTalk, Feishu, and QQ, and can even run offline. Its architecture is closest to the original OpenClaw, but it adds a long-term memory engine called ReMe, which understands you better the longer you use it.
NetEase Youdao LobsterAI — Completely free, open-sourced under the MIT license, with 16 built-in preset skills (document processing, data analysis, poster generation, you name it). Double-click installation on Windows and Mac with zero configuration. It is undisputedly the hottest “Lobster” for C-end users in China right now.
ByteDance pushed three products at once: UI-TARS Desktop (pure visual recognition interface, highest technical difficulty), Volcano Engine deployment solutions, and Coze Workspace Integrated Edition. Free users can directly run two instances.
Tencent WorkBuddy — Directly integrates with WeCom, Tencent Docs, and Tencent Meeting, complete with hierarchical permissions and audit logs.
Moonshot AI Kimi Claw — The most hardcore feature is called Agent Swarm, capable of running 100 sub-agents in parallel, which the company claims boosts efficiency by 4.5 times. Priced at 199 RMB per month.
The product list in this image did not exist at all two months ago.
Let’s do the math.
Claude Code, Anthropic’s official programming assistant, costs at least $20 a month, and $200 a month for heavy users. Plus, it only runs Anthropic’s own models, isn’t open-source, can’t be customized, and doesn’t allow for code reviews.
LobsterAI? $0. CoPaw? $0. The source code is entirely public. You can modify whatever you want and connect any large model you prefer—Qwen, DeepSeek, Kimi, Doubao, ERNIE Bot, take your pick.
$200 per month vs $0, and the features are far more tailored to the actual needs of Chinese users. You don’t need a calculator to figure this math out.
What’s even more thrilling is the speed. How long did it take for OpenClaw to go from a viral hit to forming a complete industry chain—foundations, big tech products, top 3 cloud one-click deployments, domestic chip adaptation, and bilingual Chinese-English communities? About two weeks. Some say the last time they saw this kind of speed was during the LLM explosion in early 2023, but it took more than half a year for the industry structure to form then. This time, it was several times faster.
The People Who Didn’t Rest During the Spring Festival
I want to write a bit more about this part.
A Wall Street Journal report mentioned that during the Spring Festival, numerous Chinese founders gave up their holidays, working overnight on new projects based on OpenClaw. At a spontaneously organized offline salon in Beijing, over 300 people crammed in to listen to entrepreneurs and engineers share their experiences. One attendee’s exact words were: “In this era, Chinese tech professionals share only one consensus—we absolutely cannot afford to fall behind.”
Mindverse, an AI startup in Hangzhou, organized a five-day online hackathon. Participants built an AI version of Tinder (agents screening dates for you), an Agent-to-Agent recruitment platform, and an app that lets an AI avatar travel the world and write travelogues. Five days, from scratch to launch.
I previously chatted with several friends in AI, and one detail left a deep impression on me: When NetEase Youdao was building LobsterAI, they built-in 16 preset skills covering Office document processing, video production, browser automation, and more. Why take this step? Because they realized that the original OpenClaw’s skills required users to write scripts themselves or download them from the community, creating a high barrier to entry for ordinary people. Youdao’s engineers leveled this barrier—double-click to install, ready out of the box.
There’s another particularly inspiring event: Loongson Technology successfully ran OpenClaw entirely on the domestically produced Loongson 3B6000M chip. What does this mean? It means that, in theory, an AI Agent solution completely independent of any US hardware or software now exists. The chip is domestic, the connected model is domestic (DeepSeek or Qwen), the OS runs Linux, and the Agent framework uses CoPaw or LobsterAI.
The entire chain, entirely made in China.
As for the “lead” over the US, we must unpack this. OpenClaw itself is indeed a great project, and Steinberger’s personal capabilities are incredibly strong. But frankly speaking, the value of an open-source project ultimately depends on who can land it, use it, and push it to the extreme. Google and Anthropic chose to ban it—what are they protecting? They are protecting the monetization models of their own Claude Code and Gemini. Meanwhile, Chinese companies chose to fully embrace it and build better localized versions—protecting the interests of users and developers.
Which choice is smarter? The market will provide the answer.
How Ordinary People Can Eat This “Lobster”
If you aren’t a programmer, what does this have to do with you?
It has everything to do with you.
First, if you are a professional, you can use it right now. Install LobsterAI with a double-click, configure an API key (the free quotas from Doubao or Kimi are sufficient), and tell it, “Every morning at 9 AM, grab the top HackerNews articles and send them to my DingTalk,” and it will actually do it. Writing weekly reports, organizing files, managing schedules, or even remotely controlling your PC when you’re away—just send a message on DingTalk from your phone.
Second, if you’re choosing an AI tool, domestic “Lobsters” can completely replace overseas solutions. And to be brutally honest, in the Chinese usage environment, the domestic ones work better than the original—because the original doesn’t support DingTalk and Feishu, which are the lifeblood for most Chinese professionals.
Third, if you follow investment or employment trends, this AI Agent wave is driving more than just the software layer. Cloud server orders are surging, domestic chip adaptations are accelerating, and hardware vendors (like NAS devices, mini PCs) are seeing a visible growth in sales due to the demand for “raising lobsters.” Analysts estimate that by 2030, the global commercial scale of AI Agent interventions will reach 3 to 5 trillion dollars.
If you want to support domestic products, LobsterAI and CoPaw are both completely free and open-source. It doesn’t hurt to install and try one out. The OpenClaw version in Coze Workspace also offers a free quota. You don’t need to know how to code, nor do you need to mess around with command lines. In 2026, domestic AI tools have truly achieved three words: “Functional, easy-to-use, and free.”
It’s still cloudy outside. I just checked, and OpenClaw’s GitHub stars have gone up again. But what I find more interesting is not the number itself.
It’s that a significant portion of those 250,000 stars comes from Chinese developers. It’s the 30+ domestic variants that went from zero to one in two months. It’s the nearly thousand people lining up outside Tencent’s headquarters to install the lobster. It’s the moment when it successfully ran on the Loongson chip.
The US created the lobster, but China fed and grew the entire lobster industry. This fact, if spoken out loud, is probably more convincing than any number on a PowerPoint slide.
References：
- 60 Days of the OpenClaw Boom: “Another Collective Evolution” in China’s AI Industry Application
- Foreign Media Marvels: Hotter Than Silicon Valley! OpenClaw Sweeps the Chinese AI Circle
- Panoramic Comparison of OpenClaw Alternatives: In-depth Review of 20+ AI Agent Tools in 2026
- Who is OpenClaw creator Peter Steinberger?
- OpenClaw Takes Off in China As Chinese Tech Giants Embrace the Agent
—— Lyra Celest @ Turbulence τ.
