Alibaba Cloud's price hike of up to 34% is merely the tip of the iceberg. A revolution in token consumption, triggered by the open-source AI agent framework OpenClaw, is reshaping China's AI industry landscape at a pace far exceeding market expectations.

On the afternoon of March 18, Alibaba Cloud announced on its official website that due to exploding global AI demand and rising supply chain costs, prices for its AI computing power, storage, and other products would increase by up to 34%. This includes a 5%-34% increase for computing card products like the T-Head Zhenwu 810E and a 30% increase for the CPFS (Intelligent Computing Edition) file storage product.

While the official reason cited is "exploding global AI demand and rising supply chain costs," informed sources suggest the deeper driving force is a massive surge in token calls. Alibaba Cloud's MaaS platform, Bailian, achieved its highest-ever growth rate from January to March 2026, with computing resources being prioritized for token-based services.

The origin of this token storm is an open-source AI Agent framework affectionately nicknamed "Little Lobster" by Chinese netizens—OpenClaw. As of March 15, 2026, it has garnered over 315,000 stars on GitHub and ignited an unprecedented collective effort termed "ChinaClaw" within the Chinese internet ecosystem.

On the 17th, UBS and Barclays released thematic research reports, both converging on the same core conclusion: intensifying computing power shortages, surging demand for cloud services, and the imminent monetization of enterprise AI SaaS models will make core Chinese tech stocks like Tencent, Alibaba, and Baidu the most direct beneficiaries.

The "Little Lobster" Sweeps China: From Developer Toy to National AI Infrastructure

OpenClaw's disruptive nature lies in its architectural logic—it is not merely a chatbot that answers questions in the cloud, but an "AI executor" that resides within instant messaging software and can directly control a computer to perform real-world tasks.

Its ecosystem core is the ClawHub skill platform, a repository akin to an "App Store for AI Agents," which has accumulated over 20,000 skills covering scenarios like file management, browser operations, email handling, and code generation.

More critically, a single OpenClaw instance can generate hundreds or even thousands of model calls per day, resulting in token consumption that is an order of magnitude higher than traditional AI chat applications.

This characteristic has directly fueled a rapid expansion of the global AI model API market. Data from the API aggregation platform OpenRouter shows recent weekly total token usage has reached approximately 16 trillion, roughly triple the level seen in January 2026, before OpenClaw's release.

Compared to Western markets, the adoption of "Little Lobster" in China has taken a distinctly different form. While it remains largely within niche technical developer circles in the West (involving complex setups requiring Docker environments and API key management), Chinese internet giants have rapidly repackaged it into "one-click deployment" cloud services and rolled them out en masse to both enterprise and consumer markets.

ChinaClaw in Full Bloom: Giants Place Collective Bets, SaaS Monetization Emerges

From late February to March 2026, major Chinese internet companies and AI model firms intensively launched their own "Claw" products.

Tencent's WorkBuddy was already used internally by over 2,000 employees before its public launch, focusing on data analysis and office automation scenarios for HR, administration, and operations. These products typically support invocation through mainstream work communication platforms like WeChat Work, DingTalk, and Feishu, directly integrating into enterprise workflows.

Token Explosion: Data Reveals Global AI Demand Inflection Point

The proliferation of OpenClaw is triggering an unprecedented revolution in token consumption.

UBS reports indicate that weekly AI model token usage monitored via the OpenRouter platform has reached approximately 16 trillion in recent weeks, nearly triple the level observed in January 2026, prior to OpenClaw's debut.

The underlying logic for this explosive growth is clear: compared to standard AI chat tools, the token consumption generated by using Claw agents is an order of magnitude higher. A single OpenClaw can produce hundreds to thousands of model calls daily, far exceeding the token volume of standard chat interactions.

Chinese models are the biggest beneficiaries of this token boom. According to OpenRouter data for the month leading up to March 9, 2026:

- MiniMax M2.5: Token usage reached 7.5 trillion, ranking first globally. - Kimi K2.5: 4.2 trillion, ranking second. - DeepSeek V3.2: 3.3 trillion. - GLM-5: 2.2 trillion.

Vendor-specific data also shows staggering growth rates. MiniMax disclosed that the daily token consumption for its M2 series text models in February 2026 surged over sixfold compared to December 2025, with token usage for coding plans increasing more than tenfold. The MiniMax M2.5 model, released on February 11, 2026, topped OpenRouter's popularity chart within 12 hours. Its weekly token usage jumped to 3 trillion within a week of release, exceeding the combined total of Kimi K2.5, GLM-5, and DeepSeek V3.2 at the time.

Data from Moonshot AI (Kimi) is even more striking: revenue generated by Kimi K2.5 in the 20 days following its release at the end of January 2026 surpassed its total revenue for the entire year of 2025, with overseas revenue exceeding domestic revenue. Furthermore, OpenClaw designated Kimi K2.5 as its first official free core model.

The Chinese Models' Cost "Nuclear Weapon": Priced Under 10% of US Counterparts

Exploding token consumption, combined with the significantly lower pricing of Chinese AI models compared to their US counterparts, is reshaping global developers' model selection logic.

According to Barclays Research data, as of March 9, 2026, the output pricing of mainstream Chinese AI models is generally below 20% of OpenAI's GPT-5.4, with most being under 10%.

For example, running MiniMax M2.5 continuously for one hour at an output rate of 100 tokens per second costs only about $1, merely 1/10th to 1/20th the cost of using Claude Sonnet for equivalent usage. This price advantage means that even if users initially build their Claw agents using Claude or GPT, they can easily switch to Chinese large models to control costs when the expense of using leading US models becomes prohibitive.

Notably, despite significant price reductions, the inference gross margins for Chinese AI vendors continue to improve.

MiniMax disclosed that its per-token inference cost decreased by over 50% between December 2025 and February 2026, while pricing remained unchanged during the same period. UBS research indicates that API and enterprise AI deployment gross margins for Zhipu AI and MiniMax are approximately 60%-70%, primarily derived from their inference businesses.

From "Free" to "Subscription": China's SaaS Awakening Moment

Perhaps what excites capital markets most about ChinaClaw is not just the surge in token consumption, but its role in prying open the historic gateway to SaaS-style monetization in China.

For a long time, consumer-grade AI chat products in China were almost entirely free. The introduction of ChinaClaw has brought subscription-based pricing models to both ordinary users and the enterprise market—with monthly fees typically ranging from 40 to 500 RMB, often accompanied by usage-based point packages.

Barclays Research characterizes this phenomenon as "China's SaaS Moment"—suggesting that China, which never truly developed a substantial traditional SaaS industry, might leapfrog directly into a new era of AI-driven productivity subscription services, propelled by the wave of AI and ChinaClaw.

Intensifying Computing Power Shortages: Who Are the Most Certain Beneficiaries?

The imbalance between computing power supply and demand is worsening rapidly under the impact of the ChinaClaw wave.

Barclays channel checks indicate that both Tencent and Baidu are considering raising token prices, following Alibaba Cloud's lead. More severe computing power shortages imply that larger-scale capital expenditure investments are inevitable.

Synthesizing analytical frameworks from UBS and Barclays, the impact of the "Little Lobster" frenzy on China's AI industry chain can be broken down into three tiers:

Infrastructure Layer: Most Certain Beneficiaries

- Alibaba: The largest domestic cloud provider, possessing full-stack AI capabilities. As AI computing power shortages intensify abruptly, Barclays expects Alibaba to further increase its capital expenditure guidance during its earnings release. - Kingsoft Cloud: AI workloads account for over 30% of total revenue, with deep collaborations established with several major AI labs; UBS assigns a Buy rating.

Large Model Layer: Direct Monetization of Token Boom

Chinese AI labs are the direct beneficiaries of the surge in token usage. Their cost advantage is attracting global developers to switch to Chinese models, especially as ChinaClaw accelerates adoption and token consumption per task grows exponentially. The logic of rapid growth in model API revenue has been validated by the data.

Application/Platform Layer: Most Anticipated SaaS Monetization Paths

- Tencent: Its combination of QClaw (consumer edition) and WorkBuddy (enterprise edition) covers the widest instant messaging entry points in China via WeChat, QQ, and WeChat Work, with continuous improvements in AI agent development execution. - Baidu: Baidu Search ranks first on the ClawHub skill download chart, with over 36,000 downloads. Strong demand for native search skills is driving proactive renewals of cloud service contracts.

Barclays concludes that AI computing power shortages have intensified sharply due to the emergence of Claw—not only because Claw represents a entirely new AI usage scenario, but also because token consumption in this scenario is an order of magnitude higher than ordinary chat. Once ChinaClaw gains a firm foothold on the enterprise side in China, it will open up a new and substantial Total Addressable Market. More importantly, this monetization opportunity is not necessarily confined to China—the significant cost advantage of Chinese AI models could potentially help Chinese tech companies compete effectively in overseas markets that are receptive to Chinese AI models.