In early March, outside the Tencent headquarters in Shenzhen, Tencent engineers set up stalls in the northern square of the building, offering free installations of the &#34;Lobster&#34; OpenClaw for users.
The line was endless, with some carrying NAS devices, others bringing MacBooks, and some lugging mini PCs, reminiscent of a geek gathering from a decade ago when people were flashing Android systems.
In fact, many major companies are intensively pushing their own &#34;Lobster.&#34;
Xiaomi has begun internal testing of MiclawAgent, hoping to embed AI agents into Xiaomi's &#34;full ecosystem of people, vehicles, and homes,&#34; allowing smartphones, cars, TVs, and home appliances to become execution nodes for AI.
Cloud vendors are starting to &#34;set up stalls,&#34; as terminal giants begin to integrate Agents into operating systems. This &#34;Lobster&#34; storm has already opened the curtain on the second half of the large model era.
This is not just a simple competition of AI tools, but a covert battle over the next generation of &#34;super entry points.&#34;
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/f4f41cc7-afd2-45dc-8c17-06f85f2e2f95.jpeg?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/f4f41cc7-afd2-45dc-8c17-06f85f2e2f95.jpeg"/>
<h2>Cash Flow from Selling Tokens</h2>
Currently, a dilemma faces all players: the pure &#34;Chat&#34; model cannot generate a healthy business model.
Over the past two years, domestic cloud vendors and tech giants have been caught in a long-term arms race, with thousands of high-end computing cards being systematically pulled into data centers. By 2026, ByteDance, Alibaba, and Tencent's combined capex will exceed $60 billion. However, if users do not engage, the computing power will be wasted, incurring high depreciation costs daily.
The reality is that relying solely on C-end user dialogue modes not only fails to consume such a massive reserve of computing power but also cannot generate revenue from users accustomed to free services.
Users occasionally ask AI to write an email or draw a picture, but the token consumption for such single interactions is low, unable to cover the depreciation and operational costs of the vast underlying computing power clusters. To get the expensive computing power working and generate real cash flow, the giants urgently need a &#34;Token black hole&#34; that can continuously and automatically consume computing power.
Local deployment Agents like OpenClaw have emerged to fill this role.
When users issue complex commands, OpenClaw breaks down tasks, searches online, calls local software, identifies errors, and self-corrects and retries. Each of these steps sends requests to cloud API interfaces. The token consumption for completing a complex task can be hundreds or even thousands of times that of a regular dialogue.
An AI analyst pointed out to Wall Street that &#34;Chinese open-source models are adopted by OpenClaw mainly due to their cost-effectiveness. Compared to overseas competitors, the lower costs allow for more frequent API calls, which directly translates into cash flow for cloud vendors, avoiding the waste of massive computing investments.&#34;
This is why cloud vendors like Tencent are willing to invest manpower to set up offline &#34;stalls&#34; to help users deploy open-source Agents, and why Alibaba strongly promotes OpenClaw for cloud deployment. Each deployment buries a 24-hour roaring &#34;computing power pump&#34; in the user's local or cloud computer
Regardless of whether the front-end is running an open-source model, as long as the inference and tool-calling APIs point to their own cloud services, a massive number of small requests will ultimately converge into considerable B2C and B2B cash flow. In the current capital market's harsh scrutiny of the commercialization of large models, this API flow driven by Agents is a key lifeline for giants to maintain their computing power expansion.
<h2>Mining Trajectory Data</h2>
Crossing the first layer of cash flow ledgers, the giants' second-layer goal of promoting local Agents touches upon the ceiling of large model development: the depletion of high-quality training data.
In recent years, the core resources in the competition for large models have been computing power and training data. However, as model capabilities continue to improve, another resource is becoming increasingly important: task trajectory data.
The current consensus is that high-quality publicly available text data on the internet (Wikipedia, news reports, books, and papers) has been largely consumed by various large models. If they continue to feed on this static text, large models will only become a more knowledgeable &#34;bookworm&#34; and will not be able to move towards true AGI.
What does the next generation of large models need? It needs to understand how humans &#34;take action&#34; in this digital world. This is the highly sought-after &#34;task trajectory data.&#34;
When users ask AI to accomplish a task, the AI goes through a series of steps. From understanding the requirements to searching for information, then calling tools, filling out forms, and completing payments, each action leaves a record. These records form a complete task chain.
For Agent models, this data is more valuable than ordinary text because it reflects the action logic in the real world.
This is precisely the data that has been most difficult for the giants to obtain. This data is hidden within countless fragmented software, closed apps, and deep within corporate intranets, and even search engines with vast crawling ecosystems are powerless.
The OpenClaw deployed on user terminals and the system-level miclaw serve as &#34;data detectors&#34; deep behind enemy lines.
Alan Feng, the China community manager for OpenClaw, pointed out: &#34;After users install OpenClaw, they often expect magical automation, but the real value lies in clearly defining tasks. The feedback from trajectory data allows the model to continuously optimize, and manufacturers can use this to enhance capabilities.&#34;
When users run Agents locally, allowing them to perform operations on their behalf, the Agent records every operational intention and software interaction trajectory of the user. The intensive promotion of Agent applications by domestic giants is essentially a distributed, unprecedented scale of data crowdsourcing.
Users think they are getting a free AI labor force, but in reality, as they guide the Agent and correct its mistakes, they are providing the giants with the highest quality reinforcement learning fine-tuning data for free.
Once this &#34;trajectory data&#34; flows back to the cloud, it will become a core barrier for large companies to train the next generation of Agent large models with strong logical reasoning and execution capabilities. This is akin to how Tesla collected real road condition data through millions of electric vehicles on the road, ultimately feeding back into its FSD autonomous driving algorithm.
**
Insiders of Alibaba's Qwen project told Wall Street News: &#34;The probability of China leading a new paradigm is less than 20%, but through agent trajectory data, Alibaba can quickly iterate models and narrow the gap.&#34;
Currently, tech giants are turning users' computers and mobile phones into &#34;data collection vehicles&#34; for the AI era. Whoever can master the most trajectory data will be the first to train a truly &#34;fully-formed&#34; super model.
From this perspective, the promotion of local Agents by major companies is not just for a new tool. They are still competing for the operational entry point of the AI era.
The Entry War Recycles
The Chinese internet has actually gone through several rounds of typical entry wars. Early portal websites competed for homepage traffic; during the search era, Baidu became the information entry point; in the mobile internet era, the user entry point shifted to Apps, with WeChat, Alipay, and Douyin gradually becoming traffic centers.
However, the emergence of AI is changing this structure.
Alibaba's Qianwen continues to invest in &#34;AI services,&#34; allowing users to place orders with just one sentence; Xiaomi is beta testing miclaw, deeply embedding it into the underlying system of mobile phones. These actions signal that in the future, the interface for user interaction with the digital world will be restructured.
When users become accustomed to expressing their needs in a single sentence, the operational path will change. Users will no longer actively open a specific App but will delegate tasks to AI. AI will decide which platform to use, which service to call, and which payment chain to complete.
Therefore, in such a system, the status of Apps will change. They will still exist but will increasingly become service nodes. The real entry point is the Agent that helps users complete tasks.
In this new context, &#34;competing for App entry&#34; has become an outdated concept. The real war is to become the &#34;underlying agent&#34; that directly obeys the user and controls the overall situation.
If a giant can make its Agent dominate the user's terminal, it will hold the highest power in the business world—the power of intent distribution. It can easily divert food delivery orders to its affiliated companies and direct travel needs to its payment ecosystem.
In this new &#34;walled garden&#34; built by Agents, those once invincible super Apps will be reduced to mere &#34;pipes&#34; providing basic service interfaces, completely losing the opportunity for direct dialogue with users, and even losing brand premium and traffic premium.
This is also why major companies are so sensitive to Agents. Everyone wants to become the platform that controls the Agent.
<h2>The Calm Before the Storm</h2>
The explosive popularity of OpenClaw may just be a signal.
The real change is that AI is transitioning from a &#34;talking tool&#34; to a &#34;doing system.&#34; Over the past two years, the core goal of the large model industry has been to improve intelligence levels, but now more and more companies are starting to think about another question: how to enable AI to gain action capabilities.
Once AI can reliably complete tasks, the structure of the internet will change. Many applications may retreat to the background, and users will only need to face one Agent to complete most digital life operations
In this world, an Agent acts as a new operational layer, connecting users with all services.
Looking back at the history of technology, every platform-level change often begins with something that seems insignificant. Android was initially just a system for tech enthusiasts to flash their devices, WeChat official accounts were merely a simple content tool when they first appeared, and mini-programs seemed more like lightweight web pages when they were launched.
However, these products later evolved into new platforms.
If AI truly enters the Agent era in the future, then today's OpenClaw may very well be one of the earliest names to be remembered.
What the Chinese internet is experiencing may just be the eve of this storm

TCTZF

00700

BABA

- In early March, Tencent engineers offered free installations of "OpenClaw" at their headquarters in Shenzhen, signaling a push towards integrating AI Agents across devices.  
- This initiative, along with similar efforts by companies like Xiaomi, aims to create a continuous demand for computing power by allowing users to execute complex tasks through AI, generating valuable task trajectory data.  
- The rise of local agents marks a significant shift in how users interact with digital services, potentially changing the landscape of the internet and establishing a new era where AI Agents become the primary interface for task completion.

China is stirring up an OpenClaw storm