DeepSeek is so popular, what did the state-owned central enterprises do...

Written by

Audrey Miles

Updated on:July-15th-2025

DeepSeek is a hot topic recently. When DeepSeek became the "hottest newcomer" in the field of artificial intelligence, state-owned central enterprises have already quietly laid out their plans and accelerated the wave of embracing artificial intelligence. On February 19, 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.

Positive progress of state-owned central enterprises in the field of artificial intelligence

1. The supply capacity of intelligent computing power has been significantly improved

As of the end of June 2024, the scale of intelligent computing power of central enterprises has doubled year-on-year, especially in Shanghai, Hohhot and other places, where 10,000-card clusters have been built, which have initially realized the scheduling of diversified heterogeneous computing power, supported the training and iteration of general large models of hundreds of billions and above, and provided a solid foundation for the implementation of large-scale applications.

In order to implement the "AI+ Special Action", China Telecom uses "Xingchen" as its brand and deploys in five major areas of "1+1+1+M+N", including 1 intelligent computing cloud base, 1 general large model base, 1 data base, M internal large models and N industry large models. In January 2024, China Telecom launched the country's largest operator-level intelligent computing center in Shanghai, with a computing power cluster size of up to 15,000 cards. It fully adopts self-innovated AI chips and liquid cooling technology solutions. It is currently the largest domestically produced liquid-cooled intelligent computing center with the highest single-pool training capacity.

China Mobile released the Jiutian·Zhongqing base model in October 2023, covering language, vision, voice, structured data and multimodal large models. In February 2024, China Mobile Intelligent Computing Center (Chengdu) completed the installation of the intelligent computing main server and entered the trial operation stage, introducing new intelligent computing servers and high-speed lossless network technology, and the floating-point computing capacity of a single machine increased by 50% year-on-year.

China Unicom has created the Yuanjing "1+1+M" big model system, which includes 1 basic big model, 1 big model base and M industry big models, realizing systematic empowerment that combines general capability provision with rapid customization of industry scenario models.

2. Implement high-value industry application scenarios

PetroChina, together with China Mobile, Huawei and others, has created the Kunlun Big Model, trained and released big models of different levels and types to meet the needs of different business scenarios. Its released language big models and visual big models have achieved remarkable results in industry knowledge question answering, industrial visual understanding and other aspects. By integrating massive amounts of oil and gas exploration and development data, PetroChina has established a high-quality industry data set, which plays a role in exploration and development, refining and chemical industries, and optimizes production processes through intelligent means to improve energy development efficiency.

The "Longyin Wanjie" digital productivity platform of China National Nuclear Corporation. This is the first digital productivity platform in the nuclear field in China. It combines the business scenarios of the nuclear industry to quickly design and develop digital assistants to improve work efficiency and decision-making quality. The "Longyin Wanjie" digital productivity platform combines the business scenarios of the nuclear industry to quickly design and develop digital assistants. The platform focuses on data security and autonomous controllability, and provides a high-standard example for the construction of nuclear industry data sets .

The first self-controlled large-scale "big watt" computing power base in the power industry. (Data picture)

The State Grid Corporation of China has established a sample library containing multimodal data of the power industry. Based on the nationally produced hardware computing power and algorithm framework, it has built the first autonomous and controllable cross-modal power industry professional large model "Big Watt", and applied it to the production environment in the areas of power decision-making, power image analysis, power question and answer, and power data generation.

COSCO SHIPPING's Hi-Dolphin shipping big model service platform. The platform combines shipping business needs to provide intelligent shipping management services, including route optimization, ship scheduling, cargo tracking, etc., to improve shipping efficiency and safety.

China FAW's automotive industry big model. Based on the general basic big model, China FAW has developed an independent and controllable automotive industry big model, which is mainly used in product research and development, production and manufacturing, marketing services and other fields to improve the company's operational efficiency and product competitiveness.

In short, DeepSeek promotes "AI equality" through algorithm innovation and open source strategy, breaks the traditional computing power monopoly, and shifts the global AI competition to the core competition of data scale and quality. State-owned central enterprises rely on massive high-value industry data in the fields of energy, transportation, finance, etc. (such as scenario-based data resources such as State Grid and PetroChina), release the value of data elements through technologies such as classification and grading, privacy computing, and rely on policy support and production-investment integration models (such as special funds and intelligent computing center construction) to strengthen data sovereignty. In the future, state-owned central enterprises need to use the "data-scenario-ecology" trinity strategy to transform data advantages into industrial control and build an open and inclusive AI ecosystem.

Five measures to develop the artificial intelligence industry of state-owned enterprises

1. Increase original innovation, strengthen "root technology" research and ecological coordination

State-owned central enterprises need to focus on the dual-wheel drive model of "independent control + ecological co-construction", and increase independent research and development investment around "root technologies" such as the underlying framework of large models, algorithm tool chains, and chip design. For example, China Mobile's "Jiutian Big Model" has achieved an 80% localization rate, setting a benchmark for the industry.

We will promote the joint establishment of innovation alliances between central enterprises, universities and private enterprises to form a collaborative chain of "production, learning, research and application". For example, China Southern Power Grid and Huawei have jointly built a large model for the power industry to accelerate technology iteration.

Establish pilot bases and scenario application innovation centers to shorten the cycle from laboratory results to industrialization. For example, the State Grid quickly implements technology through intelligent inspection robots.

2. Build high-value scenarios and empower the industry with data

Central enterprises should give full play to the dual advantages of "scenario + data" and focus on high-value areas to form a demonstration effect. Focus on the layout of national strategic areas such as energy and transportation. Promote inclusive scenarios such as smart cities, education, and elderly care, such as China Telecom's smart community solution.

Build high-quality industry data sets, such as the China Southern Power Grid integrating power production and user demand data to form a multimodal knowledge base. Accelerate the construction of intelligent computing centers (such as China Telecom Shanghai Liquid Cooling Center), optimize the scheduling capabilities of multi-heterogeneous computing power, and support the training of large models at the trillion-level.

3. Strengthen industrial development, coordinate planning and industrial chain collaboration

Central enterprises need to optimize their industrial layout through "top-level design + ecological synergy". In the "15th Five-Year Plan", artificial intelligence is clearly defined as the core direction, and technological breakthroughs and application promotion goals are set in stages. For example, PetroChina has made "Digital Intelligence Oil" a strategic initiative for the group's development in 2025.

Technologically superior companies (such as China Electronics Cloud) focus on chip and algorithm research and development; manufacturing companies (such as CRRC Corporation Limited) focus on the production of intelligent equipment; and scenario-rich companies (such as State Grid Corporation of China) promote the integration of technology and business.

4. Increase capital investment and combine capital operation with long-term strategy

Optimize capital allocation strategies and build a "patient capital + industry-investment linkage" model. Establish AI special funds to support long-term and high-risk basic research (such as Jiangsu's 50.6 billion yuan emerging industry fund), cultivate AI start-ups, and form a "leading enterprise + small and medium-sized innovation" echelon.

Acquire key technologies (such as AI algorithm companies) and market resources through mergers and acquisitions, shorten the path to technology commercialization, explore models such as data asset inclusion and intellectual property securitization, and revitalize the value of central enterprises' data resources.

5. Optimize talent introduction and cultivation, and build an institutional environment suitable for the development of AI

It is necessary to establish an "inclusive + professional" talent system, implement long-term assessment, allow scientific research failures, and avoid short-term performance pressures that inhibit innovation.

Pilot the chief scientist system (such as the Shandong experience) to grant technical decision-making power. Promote the deep integration of AI talents and industry experts. For example, China National Nuclear Corporation jointly develops intelligent decision-making systems with nuclear energy engineers and AI algorithm experts and builds joint laboratories with universities (such as Tsinghua University AI Research Institute) to cultivate compound talents.