On March 19, Feishu announced a brand-new upgrade to its Feishu Daily intelligent agent platform, allowing users to activate and obtain their personal Feishu-version 'Lobster' with a simple click. Paired with the AI building capabilities of multi-dimensional tables, Feishu has made another strategic move amid the OpenClaw craze.

On the other side, Tencent officially launched its customized 'Lobster' QClaw on the 20th, instantly becoming the focal point of discussion.

Pony Ma even expressed his fondness for this 'Lobster' during an earnings call, stating that the 'Lobster' application enables AI to be implemented in various rich scenarios rather than being solely concentrated in chatbots as in the past. This better leverages Tencent's resources, allowing coordinated efforts across various fronts and providing inspiration for the WeChat intelligent agent currently under development.

Alibaba also released its AI Agent platform 'Wukong' on March 17—an independent product positioned as an enterprise-level AI-native work platform, similar to OpenClaw with a safety constraint. The platform, launched as an independent application, has begun invitation-only testing and will also be directly integrated into DingTalk. 'Today, we break apart DingTalk and rebuild it with AI to forge Wukong,' said Chen Hang, CEO of DingTalk, at the launch event. 'In the past, people used DingTalk for work; in the future, AI will use DingTalk for work.'

Big players are as if by prior agreement (coincidentally) raising shrimps. OpenClaw, an open-source agent framework quietly launched on GitHub in late 2025, is rapidly reshaping perceptions in the tech circle. Mainstream cloud providers such as Tencent Cloud, Alibaba Cloud, Huawei Cloud, and Volcano Engine are competing to integrate it.

The signal is clear enough. Big players are shifting from 'selling models' to controlling the entire token ecosystem—generation, transmission, and consumption. Tokens are the 'electricity' flowing through this 'power grid.' Whoever can embed agents into the chatboxes of 1.4 billion users will define the entry point of the next era.

I

The underlying logic behind OpenClaw's popularity among big players lies in the token consumption curve.

As a project positioned as a 'personal AI assistant,' OpenClaw does not demand high hardware requirements and can be flexibly deployed in the cloud or locally, allowing the number of 'Lobsters' to proliferate rapidly. In early March, Tencent Cloud initiated a free OpenClaw installation event in Shenzhen, drawing large crowds. Within hours, hundreds of 'Lobsters' were deployed on Tencent Cloud servers.

However, regardless of where they are deployed, the operation of 'Lobsters' almost entirely relies on the cloud APIs of major model providers and is priced based on token consumption.

While ordinary AI consumes a few hundred tokens per question-and-answer session, OpenClaw can consume tens to hundreds of times more for the same task. Cases on the Zhihu community show that an unoptimized bot sending the entire chat history at once can result in over 50,000 tokens consumed. Users with improper configurations can easily see monthly bills exceed $1,000.

The surge in token consumption is directly reflected in cloud demand. According to Discovery Report data, China's overall daily token consumption was 100 billion in early 2024, surpassed 30 trillion by mid-2025, and reached 180 trillion in February 2026 at the daily level for mainstream large models. This represents a 180-fold increase over two years. According to 21st Century Business Herald, 90% of token consumption tasks come from 'raising shrimps.'

OpenClaw's nature as a major token consumer, combined with its cloud deployment needs, presents an opportunity tailored for big players. OpenClaw provides an entry point—it consumes large amounts of tokens and requires the infrastructure support of big players.

Big players realize that whoever can embed agents into users' daily lives and work scenarios will define the flow of tokens.

II

Alibaba has made the most decisive pivot.

On March 16, Alibaba completed a key organizational restructuring internally. The newly established Alibaba Token Hub (ATH) business group is directly led by CEO Wu Yongming. This business group operates on par with Alibaba Cloud Intelligence and the e-commerce business group, integrating core AI businesses such as Tongyi Lab, Bailian Platform, and Tongyi Qianwen.

Wu Yongming set a clear AI commercialization goal during the conference call: Over the next five years, annual revenue from cloud and AI commercialization will surge from the 100 billion yuan already achieved this year to $100 billion, implying a compound annual growth rate of over 40%.

This marks Alibaba's most significant organizational transformation since integrating its e-commerce business groups. Alibaba's AI strategy has shifted from decentralized exploration to unified operations.

In an internal announcement, Alibaba redefined tokens from a technical unit of measurement to a strategic core asset. The core goal of the ATH business group is explicitly stated as 'creating tokens, delivering tokens, and applying tokens.'

Actions targeting OpenClaw's demand side were launched simultaneously around this goal. Alibaba Cloud's developer community uploaded multiple OpenClaw deployment tutorials, offering a 'zero-code, no command line' one-click deployment solution. Lightweight application servers paired with OpenClaw images allow users to connect by merely configuring API keys. In terms of pricing, some configurations are available for a limited-time flash sale at 9.9 yuan/month or as low as 68 yuan/year.

DingTalk simultaneously opened API call permissions. According to Alibaba Cloud's announcement, DingTalk users can enjoy unlimited API call quotas until the end of March 2026. This policy directly reduces the cost of 'raising shrimps,' converting token consumption that might have flowed to local inference into cloud leasing and enterprise service revenue.

On March 17, Alibaba released the world's first enterprise-level AI-native work platform—'Wukong.' As an independent application, it will be directly integrated into DingTalk, which boasts over 20 million enterprise organizations—a key variable distinguishing Alibaba from other big players.

The design logic of 'Wukong' clearly targets the B-end market. Unlike consumer-grade AI assistants, it aims at enterprise workflow automation from the outset. An Alibaba insider revealed that the goal is to enable every team and company to have a 24/7 'Lobster army.'

The open-source framework and big players' strategies form a closed loop here. OpenClaw amplifies token demand, while Alibaba uniformly creates tokens (Tongyi Qianwen large model), delivers tokens (Alibaba Cloud), and applies tokens (DingTalk/Wukong) through ATH, constructing a complete infrastructure system similar to a 'power grid'—even if users deploy the framework locally, inference still relies on cloud APIs; when enterprises use 'Wukong,' data and computing power will also settle within the Alibaba ecosystem.

The core of this model lies in controlling token flow. By elevating tokens to a strategic core, Alibaba is essentially redefining the business model in the AI era: extending from computing power leasing to full lifecycle token management.

III

Big players are attempting to attract users to the AI ecosystem 'power grid' by freely distributing OpenClaw as an 'electrical appliance.' Tencent, being the closest to users, finds it most convenient to distribute 'electrical appliance experience vouchers.'

As an instant messaging giant, Tencent first recognizes that OpenClaw signifies a major shift in AI from a 'chat' paradigm to an 'execution' paradigm. The core of this transformation is AI evolving from a content generation tool to a production execution entity.

Therefore, Tencent swiftly opened its core IM to agents—the customized QClaw. In addition to QQ and Enterprise WeChat, it even unusually integrated personal WeChat via mini-programs, allowing users to achieve remote control through direct WeChat conversations. It also launched OpenClaw-based tools supporting audio messages and image commands for computers to execute multi-step automated workflows.

On the enterprise side, Tencent's WorkBuddy targets B-end users. This all-scenario AI agent is fully compatible with the OpenClaw ecosystem, seamlessly integrating with Enterprise WeChat/QQ/Feishu/DingTalk. It also provides privatized knowledge bases and backend automation monitoring for enterprise users. Tencent stated that within a week of its launch, WorkBuddy's actual usage surged, triggering a tenfold capacity expansion due to service pressure.

Tencent's logic is clear: WeChat and Enterprise WeChat provide 1.4 billion-level command entry points, Tencent Cloud CodeBuddy and Lighthouse supply computing power and security hosting, and PC Manager ensures local execution and permission interception, forming a production closed loop of 'IM command—local/cloud execution—result feedback.'

The IM entry points accumulate user behavior data, cloud facilities record execution logs, and model iterations rely on scenario feedback. In this closed loop, data continuously feeds back into model optimization, forming a positive cycle.

ByteDance's logic is similar, but its advantage lies in the massive traffic from apps like Douyin and Doubao. Lowering barriers is the primary choice to convert traffic into actual users.

On March 19, Feishu announced a brand-new upgrade to its Feishu Daily intelligent agent platform, allowing users to activate and obtain their personal Feishu-version 'Lobster' with a simple click, lowering the barrier to 'click and get.'

ByteDance's ArkClaw cloud SaaS version, powered by Doubao 2.0 large model, is directly embedded into core businesses such as Feishu, Doubao App, Douyin/Volcano Engine, forming a production closed loop of 'command entry point + cloud execution + security hosting.'

ArkClaw deeply integrates with Feishu's office suite and can also call exclusive plugins from Douyin and Toutiao, gaining a unique advantage in content marketing scenarios.

For enterprise-level solutions, ByteDance simultaneously launched ByteClaw Enterprise Edition, building unified identity authentication and permission management based on ArkClaw. The Volcano Engine AgentKit platform provides a one-stop intelligent agent foundation, allowing enterprises to quickly transform internal business processes into exclusive agents, covering vertical scenarios such as data analysis, marketing, and user research.

Within ByteDance's ecosystem, Feishu/Doubao provide massive command entry points and user scenarios, Volcano Engine supplies cloud computing power, AgentKit infrastructure, and security hosting, and the Doubao 2.0 model handles core execution and multimodal understanding. Together, they upgrade agents from standalone tools to business infrastructure.

Alibaba dominates cloud infrastructure and DingTalk's enterprise channels, Tencent holds WeChat's social entry point and Enterprise WeChat, and ByteDance controls Feishu's office scenarios and Douyin's content ecosystem. The essence of the tripartite contest is emerging: leveraging OpenClaw to attract users and bind them to their respective 'power grids.'

IV

The competitive logic among AI big players has fundamentally shifted when facing this 'Lobster.'

In its strategy report 'The HALO Effect: Heavy Assets, Low Obsolescence in the AI Era' released on February 24 this year, Goldman Sachs proposed an investment framework of 'heavy assets, low obsolescence.' The report pointed out that capital in the AI era is flowing toward physical assets with infrastructure attributes, such as electricity, computing power, and data centers.

Goldman Sachs strategists named this logic the 'HALO effect.' Investors are turning to companies with 'heavy assets, low obsolescence' characteristics to mitigate AI disruption risks.

Vying for control over the entire token supply chain is a concrete manifestation of this logic in the current 'Lobster craze' competition among big players.

In the past, cloud providers sold computing power and model APIs, charging based on call frequency. Now, Alibaba, Tencent, and ByteDance are embedding tokens into super apps and enterprise workflows. The core of this contest is control over the entire token supply chain—from generation and transmission to consumption.

Whoever controls the entry points and pricing power will define the production relations in the AI era. By attracting users through free installations and low-cost deployments, big players are essentially converting token consumption into long-term revenue, attempting to build a token infrastructure system similar to a 'power grid.'

The war has just begun.