OpenClaw demand surges, with more than 10 companies taking the lead in "raising lobsters"! Is DeepSeek V4 coming? The high-quality concept stocks favored by investors preparing for funding are revealed.

Good news for AI large models.

Last Friday (March 6), the Wind DeepSeek concept index surged at the close, rising from a slight decline at the open to close up 1.87%. Among concept stocks, Tuowei Information hit the daily limit, Meiri Interactive rose over 5%, and UCloud-W increased by 4.5%.

AI Intelligent Agent OpenClaw Exploses

More than 10 companies deploy first

The sharp rise of Tuowei Information and UCloud-W may be related to the recent popularity of the AI agent OpenClaw (formerly “Clawdbot”).

Last Friday, nearly a thousand developers and AI enthusiasts gathered at Tencent Building, completing the cloud installation of OpenClaw with the help of Tencent Cloud engineers. Tencent Cloud officials said that this event used Tencent LightSpeed Cloud Lighthouse’s one-click deployment, which takes only 5 minutes for free installation, greatly reducing difficulty.

Public reports show that, according to Meituan data, search volume related to “OpenClaw” has increased more than threefold compared to the previous period. Meituan, in partnership with Lenovo Biying, launched the OpenClaw remote deployment service, allowing consumers to search “Lobster Installation” in the Meituan app to reach the service page directly.

Known as “Little Lobster” in the industry, OpenClaw is not a traditional aquatic product but a highly popular open-source AI agent, focusing on “taking over computers and freeing hands” with automation capabilities.

The Shenzhen Longgang District Artificial Intelligence (Robotics) Office recently solicited opinions on the “Measures to Support the Development of OpenClaw & OPC in Longgang District of Shenzhen (Draft for Comments).” It encourages market-oriented, professional platform carriers to launch “Lobster Service Zones,” providing free deployment services for OpenClaw, with subsidies for qualified entities; supports the development and promotion of OpenClaw-like intelligent agent tools. Contributions of key code to international mainstream communities, development of skills packages related to Longgang’s advantageous industries on skill trading platforms, and projects integrating with embodied intelligent devices may receive up to 2 million yuan in subsidies after verification.

Dongwu Securities pointed out that the new generation AI agent represented by OpenClaw has completed a fundamental leap from “question-answering tools” to “practical assistants capable of implementation,” capable of deeply embedding into business workflows and adapting to multiple scenarios, bringing disruptive efficiency improvements to research, office work, and other fields. Industry insiders believe that the rise of OpenClaw is pushing AI into a new era of intelligent agents.

In the A-share market, according to publicly available information, more than 10 companies have deployed or completed adaptation of OpenClaw.

Based on the March 5 financing balance and the changes expected by the end of 2025, among these companies, three saw financing balances increase by over 30%, namely UCloud-W, Beixin Yuan, and Loongson Zhongke.

Additionally, Shunwang Technology, Tuowei Information, iFlytek, and Qingyun Technology-U also saw financing balance increases of over 10%.

UCloud-W stated that in late January, it launched related images for deploying OpenClaw, becoming one of the first industry players to achieve cloud deployment.

Beixin Yuan’s self-developed Xinyuan MiXin product has completed adaptation and connection with OpenClaw.

Loongson Zhongke announced that hardware based on the fully domestically produced Loongson 3B6000M chip has successfully completed local deployment of the popular AI agent OpenClaw.

Furthermore, cybersecurity risks are also worth noting. Recently, the Ministry of Industry and Information Technology’s cybersecurity threat and vulnerability information sharing platform detected that some instances of the open-source AI agent OpenClaw, under default or improper configurations, pose high security risks, which could easily lead to network attacks and information leaks.

It is recommended that relevant units and users thoroughly check exposure to public networks, permission settings, and credential management when deploying and using OpenClaw, close unnecessary public access, improve security mechanisms such as identity authentication, access control, data encryption, and security audits, and stay updated with official security notices and reinforcement suggestions to prevent potential cybersecurity risks.

DeepSeek to Release Major New Model

There is also news about DeepSeek’s new large model. According to media reports, DeepSeek will release its latest large model V4, which natively supports image, video, and text generation functions. This will be the first major update since the launch of the R1 inference model in January 2025. The V4 model is expected to further meet the domestic market demand for low-cost, open-source models.

Since the release of R1 in early 2025, DeepSeek has established itself as a benchmark for open-source models globally, maintaining steady updates and iterations, and publishing important research results in international authoritative journals, especially in the fields of large model inference.

In September 2025, DeepSeek officially released the V3.2-Exp model, which introduced DeepSeek Sparse Attention (a sparse attention mechanism), exploring and validating optimizations for training and inference efficiency on long texts.

In December 2025, the company released DeepSeek-V3.2 and DeepSeek-V3.2-Speciale. The former aims to balance inference ability and output length for daily use, while the latter pushes the inference capabilities of open-source models to the limit, exploring the boundaries of model capabilities.

On January 27, this year, DeepSeek released a new document recognition model, DeepSeek-OCR2, achieving breakthroughs in OCR with the innovative “Causal Flow” visual reasoning logic.

Currently, according to OpenRouter, the largest API aggregation platform for large models worldwide, DeepSeek V3.2 ranks among the top ten in token consumption. Previously, a report titled “Token Going Global or Facing Opportunities! Kimi+DeepSeek+Zhipu+MiniMax, Concept Stocks with Concentrated Chips” noted that in February, DeepSeek’s token consumption reached 3 trillion, ranking 4th among large models.

(Source: Data Treasure)

Guosheng Securities expressed optimism about the application of productivity agents and multi-modal entertainment applications in 2026, expecting that application-driven token consumption and the AI export opportunities will accelerate.

Zhongyuan Securities noted that DeepSeek is more open and transparent, reflecting confidence in its research results, and industry expectations for the release of V4 are high. They also believe that large model costs are expected to decrease significantly, alleviating the current chip shortage domestically. If DeepSeek V4 continues to be open-sourced, its enhanced model capabilities will support downstream AI application vendors, benefiting AI deployment. V4 is also expected to bring milestone technological breakthroughs, opening a new paradigm for large model development, and accelerating progress toward AGI. Additionally, large models may deeply integrate with domestically produced chips, with training potentially fully or partially conducted using domestic chips.

DeepSeek Concept Stocks Outperform the Shanghai Composite Index

The 2026 government work report emphasized deepening the “AI+” expansion, promoting the rapid adoption of next-generation intelligent terminals and agents, and scaling AI commercialization in key industries, fostering new intelligent business formats and models.

In the view of institutions, as large models enter the era of intelligent agent execution, major economies worldwide are accelerating the construction of AI governance systems. The convergence of AI application deployment and policy support is speeding up, pushing large models from “wild growth” to “regulated development,” with technological paradigm shifts and commercial value transformation progressing in tandem.

CCID Consulting pointed out that the large model market will continue to grow explosively. In 2025, China’s large model market size was about 49 billion yuan, expected to exceed 70 billion yuan in 2026, and reach 100 billion yuan in 2027, mainly driven by MaaS services, industry-specific models, and intelligent agent applications.

According to Wind data, as of March 6, the 45 DeepSeek concept stocks in A-shares have an average increase of over 6% this year, outperforming the Shanghai Composite Index. Fourteen stocks have increased by more than 10%, with Zoomlion, Capital Online, Meiri Interactive, and UCloud-W each rising over 34%.

Zoomlion has gained over 50% since the start of the year. The company early adopted and updated its base large model, leveraging its excellent inference ability and openness, significantly enhancing logical reasoning and reducing operational costs, injecting new vitality into industrial intelligence.

Meiri Interactive has increased nearly 35% this year. Its GAI Station, an intelligent work platform, integrates deeply with large language models like DeepSeek, promoting multi-scenario applications.

UCloud-W has also increased nearly 35%. Last year, the company announced completion of adaptation for the entire DeepSeek series, utilizing GPU nodes domestically and abroad, and providing full open-source services via the UModelverse platform.

7 High-Quality Stocks Attracting Financing

From the perspective of financing, the total financing balance of the 45 DeepSeek concept stocks on March 5 was 62.275 billion yuan, up nearly 4% from the end of 2025; half of the stocks with margin trading saw increased positions, with Kingsoft Office, Meiri Interactive, and Tuoer Si each increasing their financing balances by over 50% since the end of 2025.

Further performance analysis shows that only 7 stocks had a financing balance increase of at least 3% from the end of 2025 and achieved profitability with year-on-year growth (including turnaround).

In terms of financing balance growth, Kingsoft Office, Parnian Technology, Sangfor, and Tuowei Information each increased by over 10% on March 5.

Kingsoft Office’s financing balance increased by over 55% compared to the end of 2025. The company achieved a net profit of 1.843 billion yuan in 2025, up 12.03%. In September 2025, it announced that WPS AI had integrated with the DeepSeek series models.

Parnian Technology’s financing balance increased by over 45%. As a leading independent supercomputing service provider and general supercomputing cloud service provider in China, its net profit in 2025 grew by over 80%. The company’s intelligent computing cloud platform has deployed multiple versions of DeepSeekR1, V3, V3.1, and V3.2, as well as mainstream large models like Zhipu and Qwen.

Tuowei Information’s financing balance increased nearly 15% since the end of 2025. The company turned profitable in 2025. Early in 2025, it announced that as a partner of Ascend, its “Zhaohan” series AI servers and related products had fully adapted to DeepSeek large models, building a secure and reliable computing foundation for domestic AI large models.

Additionally, iFlytek and Desay Worry each saw over 8% growth in financing balances since the end of 2025.

Desay Worry has been visited by 20 institutions this year, and has established strategic cooperation with several well-known robot companies, officially launching the robot intelligent base AICube, an AI computing terminal for robots, supporting vehicle-grade redundancy design for autonomous driving domain controllers in robot scenarios.