"Lobster" frenzy: those selling "feed" are the first to profit; institutions say token consumption is experiencing an exponential surge

Behind the explosive popularity of the AI (artificial intelligence) assistant “Lobster,” who will be the biggest beneficiary?

Recently, as the fastest-growing open-source software project in GitHub’s history, OpenClaw (commonly known as “Lobster,” formerly Clawdbot, Moltbot) has spread from Silicon Valley to streets across China. Born in November last year, “Lobster” can be deployed on virtual private servers (VPS) or local devices like personal computers. Once granted system permissions, it can directly access the user’s terminal and read local files.

Around the Chinese New Year, major companies like ByteDance’s Volcano Engine, Alibaba Cloud, and Tencent Cloud fully opened cloud services for running OpenClaw. On March 6, nearly a thousand developers and AI enthusiasts gathered at Tencent Tower, completing the cloud deployment of OpenClaw with the help of Tencent cloud engineers.

An employee of a large tech company told The Paper that colleagues have been talking about “raising Lobster” recently, and managers also suggest everyone try it out—to understand what current popular products can do.

In response to this latest AI wave, the capital market is also reacting. On March 9, “Lobster” concept stocks surged against the trend. Notably, computing power leasing and cloud computing sectors were active, with UCloud (688158.SH), QingCloud (688316.SH), and Shunwang Technology (300113.SZ) hitting 20% daily limit-ups; TOWAY Information (002261.SZ), Meiliyun (000815.SZ), YunSai ZhiLian (600602.SH), and China Great Wall (000066.SZ) also hit daily limits; Capital Online (300846.SZ) and Wangsu Technology (300017.SZ) rose over 10%.

Institutions Say Token Consumption Is Experiencing Exponential Growth

Recently, multiple listed companies have responded on interactive platforms regarding their business plans related to “Lobster.”

Cloud service provider UCloud (688158.SH) stated that in late January, it launched images for deploying OpenClaw (Clawdbot), and recently introduced ready-to-use OpenClaw lightweight cloud hosts. Chip design and R&D company Rockchip (603893.SH), intelligent device platform provider Thunder Software Technology (300496.SZ), cybersecurity firm BeixinYuan (300352.SZ), and internet tech company Shunwang Technology (300113.SZ) also announced they have released products compatible with OpenClaw.

Most analysts agree that OpenClaw will promote the application of AI Agents, significantly increase model token consumption, and drive demand for computing power.

Specifically, Huaxi Securities pointed out that as one of the most attention-grabbing projects today, OpenClaw is expected to push the AI industry into the Agent era, bringing several impacts: Agents may become one of the earliest application scenarios; token consumption will accelerate, boosting demand for computing power and maintaining inflation in the computing chain; it will support increased demand for domestic large models and accelerate their international expansion.

OpenSource Securities believes that the popularity of OpenClaw stems from significant improvements in large model inference, multimodal, and Agent engineering capabilities. As user installations grow, it could further drive token volume, inference computing power, and cloud demand. Vertical AI applications in tools, social, and entertainment sectors will benefit from continuous model capability enhancements, potentially sustaining high growth in user scale and commercial revenue (ARR).

Guosheng Securities also noted that OpenClaw has ignited a wave of intelligent agents, with token consumption experiencing exponential growth. AI Agents are moving from pilot projects to large-scale deployment, and the explosive penetration of OpenClaw confirms its commercial viability. Rising agent penetration and complexity drive token consumption surges, creating rigid demand for computing power. Mainstream vendors are already experiencing “card shortages” and “sales restrictions,” intensifying supply-demand mismatches. This drives upgrades in the computing industry chain from both technical and economic perspectives. As domestic token consumption explodes, domestic computing power providers are expected to leverage cost advantages and ecosystem development to gradually dominate infrastructure.

“Fish tank sellers and feed sellers are the first to profit”

Amid this boom, large model companies and cloud service providers are the two main direct beneficiaries of OpenClaw.

Essentially, OpenClaw is a “token digestor”—each task involves intensive calls to large language model APIs, with token consumption far exceeding traditional conversational AI chatbots. For large model companies eager for “user growth stories,” this is a golden opportunity.

According to OpenRouter data, in the week of March 2, MiniMax M2.5’s calls reached 1.87 trillion tokens, ranking first with a 15% week-over-week increase. On March 9, at the close of Hong Kong stocks, MiniMax’s stock price surged nearly 24%, closing at HKD 997—an all-time high, with intraday touching HKD 1000.

Meanwhile, although OpenClaw emphasizes local deployment to protect user privacy, for most ordinary users, purchasing cloud-based services is more practical. Major domestic cloud providers like Tencent Cloud, Alibaba Cloud, JD Cloud, and Volcano Engine quickly launched one-click deployment services for OpenClaw.

However, users who start experimenting with “Lobster” may encounter unexpected issues.

Many netizens report that OpenClaw can “act up” during tasks, performing unexpected operations or even uploading encrypted files outside the task scope to the internet. Some describe it as “using the dumbest methods to burn the most expensive money,” handling simple tasks with complex means, with all token costs borne by users—heavy users can spend thousands or even tens of thousands of yuan daily.

Additionally, as OpenClaw sweeps through the AI community, officials warn of security risks. On March 8, CCTV News issued a reminder that the Ministry of Industry and Information Technology’s cybersecurity threat and vulnerability sharing platform detected high security risks in some OpenClaw instances under default or improper configurations, which could easily lead to cyberattacks or information leaks.

Hu Yanping, a specially appointed professor at Shanghai University of Finance and Economics and a researcher in intelligent technology industry and economy, told reporters that OpenClaw has practical issues, including handling simple tasks with complex methods, frequent task interruptions, unreliable memory, and high token costs relative to actual benefits. He said, “Its level and global enthusiasm do not match.”

Due to the complex setup required for local deployment, many domestic social platforms overnight appeared offering OpenClaw installation services, priced from hundreds to thousands of yuan. Like the previously popular Manus AI products, various “course-selling” accounts also seized the opportunity to promote and attract attention.

Beyond these reasons, “Lobster” faces a more serious problem: lack of real-world application scenarios. Many users, without clear programming or automation needs, find it difficult to identify suitable use cases in daily work and life after initial excitement, resulting in a situation where “they are just looking for nails with a hammer.” Some netizens jokingly say, “The real application scenario is just collecting installation fees.”

Tan Jian, associate professor at Beijing University of Posts and Telecommunications’ School of Digital Media and Design Arts, believes that today’s “Lobster” is essentially just a “shell”—it has no independent intelligence. “It’s fundamentally a universal hub, with applications plugged into the left and large model APIs plugged into the right. Cloud providers are competing for ‘servers + bandwidth + storage,’ model providers for ‘API calls.’ They are the definite winners of this ‘shrimp-raising’ craze: ‘Whether shrimp raisers make money or not, the fish tank sellers and feed sellers are the first to profit.’”

Tan further pointed out that in the long run, the big companies are not just competing over Lobster itself but over the “new intermediary layer” it represents between humans and all digital services: “In the short term, cloud and model providers are the definite winners; in the medium term, those who develop safe, user-friendly Agent products will win; in the long term, whoever integrates Agents into mobile operating systems will be the ultimate winner.”

(Article source: The Paper)