PRC Deploys DeepSeek Across Local Governments

Executive Summary:

Beijing has moved to codify its development and deployment of artificial intelligence (AI) in domestic and international settings, including at the Two Sessions and the United Nations.
At least 72 local governments across the People’s Republic of China have already deployed homegrown AI model DeepSeek, according to a domestic think tank.
Officials expect AI deployment in government to expand rapidly with the development of new models, even though such technologies have remained limited to customer and business services in the short term.
A lack of compute infrastructure and energy resources in many parts of the country constitute a bottleneck for rapid adoption of AI products to power government services.

Governments across the world have announced plans to improve the efficacy of their services by integrating artificial intelligence (AI) models into their systems. In the People’s Republic of China (PRC), officials increasingly see DeepSeek, the homegrown company behind the impressive R1 reasoning model that launched in January, as a vehicle to promote widespread AI integration across the country (China Brief, February 11, March 19).

Within two months of its unveiling, both the private and public sectors have embraced generative AI solutions, including the country’s ports as well as governments in first-tier cities such as Beijing and Shenzhen. The central government has not announced whether it will adopt DeepSeek. Local government agencies, however, are pioneering AI integration efforts, experimenting with the technology in ways that may be emulated and promoted by Beijing in the future.

PRC Seeks AI Leadership at Home and Abroad

Beijing has been at the forefront of global regulatory approaches to AI. In July 2017, the State Council started to codify and publish national AI strategies with the release of a New Generation Artificial Intelligence Development Plan (新一代人工智能发展规划). This document, which argued that data was emerging as “the number one element of economic growth” (经济增长的第一要素), outlined three milestones to be achieved by 2020, 2025, and 2030. By 2020, it envisioned that “the AI industry will have become a new important economic growth point” (人工智能产业成为新的重要经济增长点); by 2025, it aimed for “major breakthroughs in AI theory and implementation” (基础理论实现重大突破); and by 2030, it called for the PRC to “become the world’s main center for AI innovation” (成为世界主要人工智能创新中心) (State Council, July 20, 2017; China Brief, December 22, 2017). DeepSeek’s models, whose successes are based on innovations in model architecture, suggest that the plan is currently on track.

At the “Two Sessions” this year—the annual plenary meetings of the national legislative and consultative bodies—AI took center stage. The 2025 Government Work Report (政府工作报告) was the first to address “large-scale [AI] models” (大模型) and “embodied AI” (具身智能), although AI has been referenced in all such government work reports since 2018 (Xinhua, March 10). It declared that under the “AI+ initiative” (‘人工智能+’ 行动), the country will work to “better combine digital technologies with the country’s manufacturing and market strengths” (将数字技术与制造优势、市场优势更好结合起来). Plans include supporting the “extensive application of large-scale AI models and vigorously develop new-generation intelligent terminals and smart manufacturing equipment, including intelligent connected new-energy vehicles, AI-enabled phones and computers, and intelligent robots” (大模型广泛应用，大力发展智能网联新能源汽车、人工智能手机和电脑、智能机器人等新一代智能终端以及智能制造装备) (Xinhua, March 12). Coupled with its broader focus on innovation and emerging technologies (the word “innovation” (创新) appears 40 times in the report, slightly more than in previous years), the PRC government is firmly committed to promoting—and managing—AI advancement.

Figure 1: Map of DeepSeek Deployment in Government Departments Across the PRC’s Provinces and Cities

(Source: Zhiding Think Tank)

Overseas, Beijing also positions itself to lead AI governance on the global stage. In late September 2023, Foreign Minister Wang Yi (王毅) laid out a five-point action plan at the High-Level Meeting on International Cooperation on Capacity-Building of Artificial Intelligence at the United Nations. The five points encompassed promoting the AI+ Initiative, improving infrastructure connectivity and AI-related training, and ensuring data security and AI safety (United Nations, September 25, 2024; MFA, September 30, 2024).

Beijing’s focus—both domestically and overseas—on DeepSeek and AI more broadly is indicative of the importance it places on realizing its ambition of global leadership and emerging technologies. Its signal at the country’s highest annual political forum to deploy DeepSeek across the country indicates that it seeks to have “first-mover advantage” (先发优势), as it had set out as a basic principle in the 2017 AI plan.

DeepSeek Integration Begins Across Local Governments and Businesses

Public and private stakeholders have moved quickly to deploy DeepSeek’s R1 model into their services and operations. In the private sector, many companies have begun to use DeepSeek for customer service, operational streamlining, and tracking purposes. For example, Ningbo Zhoushan Port (宁波舟山港) reported using DeepSeek for its Smart Gate (智能闸口) system only three weeks after the R1 release. It claimed to be using the model to recognize and track cargo, vehicles, and customs seals (Jingzhou News, February 17). Similarly, large companies, such as leading electric vehicle manufacturer BYD (比亚迪), have started integrating DeepSeek into their products. By mid-February, BYD had announced plans and rolled out a DeepSeek-supported driver-assistance system (天神之眼) to some of its cars (BYD, February 10). Around the same time, Zhuhai Wangxin Software Development (珠海市网欣软件开发), a computer software company in the real estate sector headquartered in Guangdong’s Zhuhai city, employed DeepSeek to accelerate contract approvals and other management processes for new real estate developments (Baijiahao, February 18).

In the public domain, municipal and provincial governments have also started to deploy DeepSeek in their daily operations. According to technology-focused think tank Zhiding (至顶), 72 provincial and city governments had fully deployed DeepSeek as of early March (Zhiding, March 10; see Figure 1). These include tier one cities such as Beijing, Guangzhou, and Shenzhen, as well as others such as Nanchang, which has branded its approach “Deepseek+Government affairs” (Deepseek+政务) (Nanchang Daily, February 21; Zhiding, March 10).

Adoption varies across the country. In Guiyang, the capital of Guizhou Province, initial deployment has focused on an intelligent chatbot, a smart 12345 hotline, and an intelligent emergency dispatch system. [1] The chatbot can help with regular queries about municipal administration, while the smart 12345 hotline is designed for non-emergency phone calls, and the emergency dispatch system uses AI to deploy emergency services. At a March press conference, Liu Ning (刘宁), the Deputy General Secretary of the Guiyang Municipal Government and Deputy Director of the Guiyang Urban Operation Management Center claimed that DeepSeek can help save lives, noting that its emergency dispatch function can recommend initial first aid measures (Baijiahao, March 10).

Officials have begun to receive training on how to use DeepSeek’s technology, with staff at Shenzhen’s municipal government getting an introduction in February. The city government made sure to inform people that they had put the model through rigorous testing before deploying it in its systems and provided reassurance that a professional team would still be available around the clock alongside the AI—an indication that suspicion around AI tools remains prevalent. Various sub-agencies in Shenzhen have also taken steps to integrate DeepSeek into their daily operations. The city’s Municipal Meteorological Bureau is collaborating with Huawei to release the first “regional AI forecasting model” (区域级人工智能预报大模型), which allows for capturing more accurate data and minute meteorological details otherwise missed (Shenzhen Special Zone Daily, February 17).

Beijing’s municipal government began implementing DeepSeek in its operations on March 4, according to an official statement (Beijing Daily, March 5). This takes the form of an application that provides 24-hour consultation services for businesses, allowing people to consult an AI tool when registering enterprises and filing forms. This tool is being framed as an “AI civil servant” (AI公务员).

The speed with which officials have deployed DeepSeek, especially in tier-1 cities such as Beijing, suggests the central government may follow suit in the near future, although no formal announcements have been made at this point.

A Roadmap for the Future Deployment of AI

Western media have reported that DeepSeek is preparing to release a new R-2 model before May 2025, though the company is yet to comment on the reports (Reuters, February 25). This rapid timeline likely is due to both growing demand for and growing competition with DeepSeek’s product—in early March, AI assistant tool Monica released its beta model of Manus and Tencent launched an upgrade to its Hunyuan (混元) reasoning model series, “Hunyuan-T1” (YouTube/Manus AI, March 5; Tencent, accessed March 26). However, while advances take place at the software level and models’ inference capabilities are finetuned, fully integrating these technologies into government and other services will take time.

DeepSeek and other AI tools currently face an infrastructure problem. They rely on vast amounts of computing power, which entails the availability of abundant energy resources, the construction of data centers and cloud services, as well as the training of a skilled workforce to maintain them. The PRC has made moves in this direction, adopting the “Eastern Data Western Computing” (东数西算) plan, a multiagency national initiative to coordinate energy and cloud investment, in February 2022 (China Brief, February 28). Over 500 new data centers were announced in the subsequent two years, at least 150 of which were finished at the end of 2024. Not all the projects, however, have been led by technology companies. Instead, they include a firm that makes MSG and another that manufactures textiles. Many of these data centers are currently unused (MIT Tech Review, March 26). As models continue to improve, officials will continue to promote their alignment with government priorities and integration with government systems.

Figure 2: Framework for Countermeasures and Solutions

(Source: Zhiding Think Tank, March 11)

An additional problem is the risks inherent in rapid technological advances. For researchers at the Chinese Academy of Sciences, AI poses challenges to efficiency, accuracy, data security, and system vulnerabilities (People AI, February 24). To mitigate these issues, they argue that governments must carefully gauge the demand for AI services, weigh the economic feasibility of deploying them, reinforce data protection, and establish rigorous oversight and maintenance protocols. PRC citizens have their own concerns about AI models, though currently these are mostly related to personal use of DeepSeek. On the FAQ-style website Zhihu, some users have shared tips on dealing with common DeepSeek issues. Many report seeing the “Server busy, please try again later” (服务器繁忙，请稍后再试) error message, often caused by heavy traffic, cyber attacks, and bandwidth constraints. One also noted that DeepSeek surpassed other AI models in terms of error rates (Zhihu, February 3). In addition to personal errors in using DeepSeek, it has also experienced a variety of Distributed Denial-of-Service (DDOS) attacks, limiting its availability to users in targeted cyber attacks (S&T Daily, January 29; DeepSeek, accessed March 26). As municipal governments move to deploy DeepSeek within their administrative services, there have been no widespread reports of outages or failures caused by AI integration in these governmental services yet. Such vulnerabilities, however, may also impact government services moving forward.

Other issues are more troubling, particularly for those outside the PRC who might wish to use DeepSeek. Beijing has a history of using emerging technologies to enhance its system of surveillance and repression, as well as in military applications. There is currently no evidence to support this where DeepSeek is concerned, but President Xi Jinping’s focus at the Two Sessions on advancing military-civil integration makes this prospect highly likely (China Brief, March 15). In this way, AI adoption within the PRC’s government operations is just part of a wider story of how these technologies have the capacity to transform how the Party imposes its will on the Chinese people.

Conclusion

Despite DeepSeek’s R1 model emerging only in January 2025, within weeks, it has already been deployed in over 70 local governments and countless businesses across the PRC. State backing for “DeepSeek+Government Affairs” suggests that this is only the beginning of DeepSeek’s products being used by official systems, with their eventual endorsement at the political center a likely prospect.

In the meantime, caveats remain that will temper the speed of AI tools’ integration. The lack of sufficient infrastructure and energy resources is one bottleneck, as is the need for updated national standards for these services. Finally, successes or failures in DeepSeek’s deployment now could have an impact on uptake further down the line.