DeepSeek | Observation on the application of DeepSeek by central enterprises: “3T” features and 8 major uses

On February 19, 2025, the State-owned Assets Supervision and Administration Commission of the State Council held a meeting to deepen the deployment of the "AI+" special action of central enterprises, summarize the progress and achievements of state-owned central enterprises in developing artificial intelligence, and study and deploy the next key tasks. The meeting emphasized that state-owned central enterprises should seize the strategic window period for the development of the artificial intelligence industry, strengthen scientific and technological innovation, focus on key areas to accelerate the mastery of "root technologies", firmly tackle large models, actively participate in the construction of an open ecosystem, promote more original innovations from "0 to 1", and accelerate the transformation of achievements and industrialization development.

In the tide of the implementation of the "artificial intelligence +" strategy, DeepSeek, a leader in domestic open source models, stands out among many solutions with its outstanding engineering optimization capabilities, superb performance and affordable cost advantages.

According to public information, many central enterprises in the energy and power sectors, including State Grid, China Southern Power Grid, China National Petroleum Corporation, China Petrochemical Corporation, China National Offshore Oil Corporation, China National Energy Group, China Huadian Corporation, State Power Investment Corporation, China Huadian Corporation, China National Nuclear Corporation, and China General Nuclear Power Corporation, have announced their accession to the DeepSeek family, covering a wide range of areas from computing power to application.

In an era where data has become a key production factor, leading domestic large models such as DeepSeek, by landing in these central enterprises with massive data resources, are accelerating the implementation of AI applications and innovating business models, ushering in a new era of intelligence and efficiency.

Image source: Jimeng AI

The reasons why central enterprises access DeepSeek are multi-dimensional and intertwined:

• Policy-driven development sets the strategic tone . It is not only a concrete practice in response to the national "AI+" strategy, but also a key path for central enterprises to transform from traditional operating models to digitalization and intelligence.
• Technical performance solves the pain points of vertical scenarios . DeepSeek's technological breakthroughs in reasoning ability, engineering efficiency, and open source flexibility have solved the high-complexity needs of central enterprises in professional fields and have become the core motivation for central enterprises to choose it.
• Cost and security requirements drive domestic substitution . The open source nature of DeepSeek lowers the threshold for model tuning and deployment, which not only meets the country's requirements for independent and controllable technology in key areas, but also reduces long-term operation and maintenance costs through computing resource optimization. The "safe and controllable" technical characteristics have become the core consideration for central enterprises in the financial, energy and other fields to choose DeepSeek.
• Ecosystem co-construction is aimed at building long-term competitiveness . Central enterprises promote industry ecosystem co-construction by connecting to DeepSeek and strengthen their voice in technical standards. This not only accelerates the transformation of AI technology from the laboratory to the industry, but also enhances the technological moat of central enterprises in global competition through ecological synergy.

3T Features

According to the cooperation cases and deployment data of central enterprises in public information, DeepSeek's implementation in central enterprises can be systematically summarized into the following "3T" core principles:

Task-Specific

By deeply adapting to the needs of vertical fields, through configurable development and single-card simulation technology, we can quickly achieve scenario-based tuning of model parameters and scale, and build exclusive solutions for "thousands of enterprises and thousands of faces".

PetroChina integrates the DeepSeek-V3/R1 model with the Kunlun model with 70 billion parameters, and adds a "deep thinking" mode to support mathematical formula recognition and seismic data processing in reservoir development, shortening the R&D cycle. China Southern Power Grid upgraded the "Big Watt" model system to adapt to full-parameter models from 1.5B to 671B, embedded power production and marketing knowledge bases, and enabled non-professionals to build intelligent applications through a "drag-and-drop" interface. China Energy Group Longyuan Power deployed the DeepSeek-R1 model, combined with the RAG architecture to integrate production data and professional documents, forming a dedicated knowledge base to improve the accuracy of equipment warnings.

Trustworthy

With data security and domestic technology as the cornerstone, we will build a "security moat" exclusive to state-owned enterprises.

Full stack localization deployment: PetroChina and China Mobile cooperated to complete the localization training and adaptation from the bottom chip to the framework, achieving 100% independent control. Intranet privatization operation: CNOOC privately deployed the DeepSeek-R1 series model through the "Haineng" platform, and the core business data such as the ERP system and the CNOOC Mall were closed in the intranet to avoid the risk of leakage. Controllable computing power resources: China Mobile provided PetroChina with more than 1000P of localized computing power, building an integrated system of "ubiquitous computing power, safe and reliable".

Transferable

Through cross-scenario knowledge reuse and model distillation technology, we can achieve “one-time training, multi-domain empowerment”.

Industry knowledge migration: Sinopec uses DeepSeek-R1 to analyze hundreds of industry standards and technical specifications, build high-quality petrochemical data sets, and support reuse in scenarios such as seismic data processing and chemical research and development. Model capability sinking: China Southern Power Grid uses the "model distillation" technology to migrate the model capabilities of hundreds of billions of parameters to small-scale models, reducing the difficulty of edge equipment deployment and improving reasoning efficiency. Cross-domain ecological co-construction: After the State Grid MSC platform is connected to DeepSeek, it supports the development of more than 300 innovative tools such as marketing systems and 95598 customer service systems, and the reuse of knowledge bases.

8 major uses

According to public information, the core uses of central enterprises' access to the DeepSeek big model can be summarized into the following eight categories, covering core areas such as production, management, R&D, and security, to promote intelligent transformation and high-quality development.

1. Intelligent production control and operation and maintenance

Optimize industrial processes through AI, reduce manual intervention, improve production stability and response speed, and improve equipment operation and maintenance efficiency.

For example, the Nuclear Intelligence AI portal developed by the Nuclear Power Operation Research Institute under China National Nuclear Corporation has been successfully connected to the DeepSeek big model. Through the AI-assisted decision-making system, the risk of human error is significantly reduced, a "smart defense line" for nuclear power safety is built, and operation and maintenance costs are optimized and equipment life cycles are extended. CGN Digital Technology integrates DeepSeek with the group's existing basic big models for scenarios such as intelligent troubleshooting and maintenance guidance for equipment failures, nuclear power plant operation data analysis and safety warnings.

2. Smart Office and Knowledge Management

By privately deploying the DeepSeek model, we build an internal intelligent assistant for the enterprise, providing services such as document generation, meeting minutes compilation, and precise retrieval.

For example, the State Grid has opened up functions such as intelligent writing and meeting minutes generation through the "i State Grid" AI assistant, significantly improving semantic understanding and task execution efficiency. China Huaneng's AI assistant "Ruizhi Xiaoneng" is deeply integrated into the "iHN+" platform, empowering all aspects of energy production, operation and management. At present, combined with corporate databases such as prompt word markets and rules and regulations, basic functions such as knowledge questions and answers, official document drafting, intelligent proofreading, document interpretation, and scientific research assistance have been realized to efficiently assist daily office and management. China Energy Construction deploys the DeepSeek model to the unified construction system portal to optimize functions such as intelligent customer service and work assistants. Intelligent review of bidding documents and automatic generation of bid evaluation reports for the State Energy Group.

3. Production process optimization and efficiency improvement

Use DeepSeek's deep reasoning capabilities to optimize energy production processes, equipment operation and maintenance, and R&D efficiency.

For example, PetroChina has shortened the R&D cycle and improved the accuracy of mathematical formula recognition for reservoir development and seismic data processing by integrating the Kunlun Big Model with DeepSeek. China National Offshore Oil Corporation's "Haineng" artificial intelligence model platform has officially connected to the DeepSeek series of models, providing open services to the entire group through private deployment. It has been launched simultaneously on the "Haineng" platform web page and the "Haineng Zhiwen" of the CNOOC Mobile Cloud, realizing intelligent question-and-answer services with secure and controllable data in the intranet environment. The State Energy Group has deployed the DeepSeek-R1 model and built a new energy-specific knowledge base in combination with the RAG architecture to improve the accuracy of equipment warnings.

4. Intelligent interaction and terminal applications

Create "thinking industrial terminals" and promote the commercialization of AI.

Dongfeng Motor R&D Institute recently announced that it has completed the access to DeepSeek's full range of large models, and will gradually install them on its own brand models in the near future. Dongfeng Motor's Lantu brand smart cockpit is connected to DeepSeek, and the smart cockpit has natural language interaction and real-time processing of high-precision maps. The official website emphasizes that "the voice assistant supports multiple rounds of complex commands, such as 'navigating to the nearest charging station and making an appointment for fast charging in 15 minutes'".

5. Scientific research and technological innovation support

Accelerate the integration of scientific research data and cutting-edge technology analysis, break the "knowledge island", and accelerate the transformation of technological achievements.

For example, the Central Research Institute of State Power Investment Corporation has completed the local deployment of DeepSeek to build a "Science and Technology Intelligence Assistant", providing more than 30 researchers in the institute with services such as obtaining cutting-edge information and in-depth analysis, helping researchers to efficiently obtain cutting-edge information. The Nuclear Power Operation Research Institute of China National Nuclear Corporation combines DeepSeek with the "Nuclear Intelligence·Longling" large model to promote the intelligence of scenarios such as operation and maintenance knowledge question and answer and dispatch order generation. The Model Service Cloud MSC (Model-Service-Cloud) platform independently developed by State Grid Corporation of China (hereinafter referred to as "State Grid Information and Communication Industry Group"), a subsidiary of State Grid Corporation of China, is fully connected to the DeepSeek large model. The deep integration of the MSC platform and DeepSeek realizes key functions such as accurate interpretation of needs, ultimate interactive experience, intelligent prevention and control of defects, performance optimization strategies and automatic document generation, improves the R&D efficiency of power grid digitalization projects, enables rapid response and flexible transformation of businesses, and effectively reduces R&D costs.

6. Secure and flexible deployment

Through domestic computing power and private intranet deployment, data security is guaranteed and complex scenarios can be adapted.

For example, China Southern Power Grid upgraded its "Big Watt" model system to support flexible deployment of full-parameter models from 1.5B to 671B, and provide distributed high-concurrency reasoning services. Inner Mongolia Electric Power built its own power grid AI based on domestic computing power, and compiled the DeepSeek user manual to guide intelligent question-and-answer, document analysis and other scenario applications.

7. Market analysis and decision support

Analyze market demand and policy changes through deep reasoning capabilities.

For example, after deploying DeepSeek, Sinochem optimized the chemical product R&D process, reduced energy consumption and costs, and achieved technological breakthroughs in reservoir development and seismic data processing. Sinopec used the DeepSeek-R1 model to analyze industry standards and technical specifications, supporting the generation of supplier business analysis reports. The State Energy Group successfully completed the local deployment of the DeepSeek-R1 series of large models on the State Energy Enterprise Cloud Platform and officially launched it, and carried out in-depth adaptation work with the Low Carbon Institute's complex energy market price prediction large model.

8. Ecosystem co-construction and industry standard promotion

Promote the construction of an open AI ecosystem and technical standardization in the industry.

For example, China Southern Power Grid will actively cooperate with various AI emerging companies to explore the application of cutting-edge AI technology in power business and jointly build an open AI ecosystem in the power industry. State Grid absorbs DeepSeek's multi-stage reinforcement learning ideas, strengthens cross-disciplinary AI talent training, and creates power-specific application scenarios.