On a Friday afternoon in March, nearly 1,000 people lined up outside Tencent’s headquarters in Shenzhen to install software on their laptops. Engineers in the company’s cloud division helped deploy OpenClaw, an open source AI agent developed by Austrian programmer Peter Steinberger, to students, retirees, and office workers.
Last month, China’s leading cloud providers debuted their own versions of OpenClaw, local governments dangled subsidies to startups developing OpenClaw apps, and a cottage industry sprang up to help users install the open source framework.
Users in China are currently trying out the phrase “raise lobsters” to refer to OpenClaw’s red lobster logo. This has proven to be a pain for Chinese AI startups and could lead to a surge in usage in the future. According to HSBC, in early February, among the top nine models on the AI market OpenRouter, Chinese AI models surpassed US models for the first time in terms of share of tokens (units of data processed by AI).
The OpenClaw craze also coincides with China’s adoption of open source AI. This strategy has helped increase the lab’s reputation among the developer community and helped the model gradually penetrate global business.
What is an open claw?
Steinberger released OpenClaw on GitHub last November, and it quickly became popular among AI developers and enthusiasts. OpenClaw is what is called an “agent harness”. This is not an AI model per se. Users must choose a model from an AI company to act as the brain of the agent. But OpenClaw consists of a set of instructions for how the AI agent breaks down a goal into a series of subtasks, a protocol that allows users to connect to the various software tools the AI agent uses, and a memory feature that means the AI agent doesn’t forget what it’s ever done.
Because the OpenClaw agent runs locally on a user’s machine and connects to tools such as messaging apps, email, calendars, and other systems, users can easily ask the AI agent to do helpful things, such as regularly checking email and automatically replying to certain messages or making reservations on the user’s behalf. Steinberger, who has a long history as an entrepreneur, was later hired by OpenAI.
Over the past few weeks, China’s largest cloud providers Alibaba Cloud, Tencent Cloud, ByteDance’s Volcano Engine, JD.com, and Baidu have all adopted OpenClaw or its spinoffs. Many startups and big tech companies have also released their own “Claw” frameworks, including Tencent’s WorkBuddy, Minimax’s MaxClaw, and MoonShot’s Kim Claw.
Shenzhen’s Longgang District offered grants of up to 10 million yuan ($1.4 million) to “private enterprises”, or companies where the founder is the sole shareholder. The city of Wuxi, near Shanghai, has dangled up to 5 million yuan ($730,000) in funding for breakthroughs in robotics and industrial applications powered by OpenClaw.
These subsidies are landing in markets where users are eager to experiment with new AI. “Younger generations in Asia, especially China, are part of a high-tech adoption culture,” said Jan Wuppermann, head of service assurance, data and AI at NTT Data. luck. “There’s an idea that I often hear from my everyday Chinese friends: It’s there anyway, so why not use it?”
In the West, OpenClaw’s popularity has declined due to security concerns. AI agents can be vulnerable to “prompt injection” attacks, where a malicious attacker plants malicious instructions on a website. OpenClaw agents were tricked into uploading sensitive data, including financial information and crypto wallet keys. In other cases, agents deleted emails and code libraries.
OpenClaw builds on a strong 2026 in China’s AI sector. Almost all major AI labs in China have released updates to their open source models, including Moonshot’s Kim 2.5, Minimax’s M2.5, and Zhipu’s GLM-5. ByteDance’s new AI video generation model, Seedance 2.0, also made headlines after its debut at the 2026 Chinese New Year Gala, one of the most widely viewed television events in China.
The move to agent AI is giving some Big Tech companies an opportunity to catch up with agile AI labs. Tencent is currently working on a new AI agent that can be integrated with its ubiquitous WeChat super app. information This was reported on March 10, citing anonymous sources. Tencent’s AI efforts have so far proved less successful than rivals Alibaba and ByteDance. Tencent’s chatbot Yuanbao reportedly has just 109 million users, far fewer than ByteDance’s Doubao and its 315 million users. information.
The OpenClaw craze has helped the stock market fortunes of some Chinese AI companies. Tencent stock has risen 8.9% over the past week. MiniMax is up 27.4% since last weekend. The stock is currently up more than 600% since its IPO earlier this year.
Still, Chinese AI startups have a long road to profitability. MiniMax announced its 2025 earnings on March 2, giving investors a first look at what the AI Lab’s financials look like.
What’s the answer? expensive.
The AI startup reported a 159% increase in total revenue to $79 million. More than 70% of this revenue comes from overseas markets, showing that MiniMax is gaining traction outside of China as well. However, the company still posted a net loss of $1.8 billion, due in part to research and development costs totaling $252 million.
Still, investors don’t seem to care. At one point last week, MiniMax was worth more, even though tech giant Baidu generates $18.5 billion in 2025 revenue, more than 230 times more than MiniMax.
China’s open source goes to the world
China’s open source model is quietly, and not so quietly, starting to spread to global business. Airbnb CEO Brian Chesky raised eyebrows last year when he admitted that the company was using Alibaba’s open source Qwen model to power its customer service agents. “It’s very good. It’s fast and cheap,” he said.
Last November, the city-state’s national AI program, AI Singapore, hired Qwen to build Qwen-SEA-LION-v4, a large-scale language model optimized for Southeast Asian languages. Alibaba currently claims that the Qwen family of models has been downloaded over 1 billion times and used by over 200,000 developers.
“I can see the appeal of the open weight model,” says Jeff Walters, head of Asia Pacific technology practice at Boston Consulting Group. “The latest Frontier models may lag a little in performance, but in many situations you don’t always need the best. Sometimes ‘good enough and cheap’ is the right tool to have in your toolbox.”
Using open source also gives companies choice and doesn’t lock them into one provider. This may be beneficial for startups trying to navigate an ever-changing world of regulations, export controls, and changing partnerships.
Still, the open source model shifts the burden of performing the computing to the user. “While it can be exciting to compare the cost per token of commercial and open source models, that’s only part of the cost,” Walters cautions.
Businesses have to pay for their own processors, but there are hidden costs as well. Wuppermann says, “Hidden costs such as security breaches and complexity are often not measured, but instead show up in other dimensions, such as additional headcount or increased time to market.”
For Wuppermann, the decision to go open source was largely a philosophical one. “People who convert to open source always insist on open source.”
China’s AI challenges
While OpenClaw and China’s open source model gain momentum, China’s AI ecosystem faces increased scrutiny over data security, intellectual property, and the Chinese government’s own shifting priorities.
In February, Anthropic accused three Chinese companies (DeepSeek, Moonshot AI, and MiniMax) of trying to extract knowledge from its cloud models. OpenAI also accused Chinese research institutes of conducting distillation attacks and using US models to train Chinese research institutes.
Oddly enough, this complaint may have boosted the reputation of the Chinese institute. Reaction to Anthropic’s accusations on social media was mixed, with some users pointing out that even if DeepSeek and others engaged in “illegal” distillation, they at least shared the results, unlike Anthropic, which has kept its AI models closed-sourced.
China’s own open source efforts may also be fraying. On March 3, Lin Junyang, the technical director of Alibaba’s Qwen model and the driving force behind the company’s open source strategy, abruptly announced his resignation.
Lin’s departure exposed tensions between Alibaba’s open source ambitions and its push to commercialize its flagship models. According to local media reports, the Qwen team disagreed with Alibaba management’s goals and expressed dissatisfaction with cloud customers sometimes getting access to computing before they could. (Alibaba asserted it has not abandoned its open source strategy)
The Chinese government may also try to dampen its enthusiasm for OpenClaw. Bloomberg reported Wednesday that both government agencies and state-owned companies have been warned not to install OpenClaw on work devices due to security risks.
Still, Chinese companies continue to release their own versions of OpenClaw. On March 12, Sensetime, once one of China’s most prominent AI companies, announced that it has integrated its office assistant Office Raccoon with OpenClaw.
And local Chinese are finding ways to take advantage of this trend. The engineers discovered a new business charging 500 yuan ($72) to install OpenClaw onsite. And what if someone gets callous about giving an AI agent access to their entire life? They’ll also charge you for uninstallation.
