China’s DeepSeek and its Open-Source AI Models
China’s pioneering start-up DeepSeek has made significant strides in the global AI race with its resource-efficient, open-source models. Despite US export restrictions on critical hardware, DeepSeek has developed competitive AI systems like the DeepSeek R1, which rival industry leaders such as OpenAI, while offering an alternate approach to AI innovation. Its success is reshaping global tech dynamics and highlighting China’s growing influence in the AI sector.
China’s artificial intelligence (AI) landscape has witnessed a ground-breaking development that is reshaping global perceptions of innovation and competitiveness. DeepSeek, a Chinese AI start-up, has stunned the tech world with its resource-efficient approach and a cutting-edge R1 AI model.
Operating under restrictions from US semiconductor export controls, the Hangzhou-based firm has achieved what many thought improbable—building a competitive large language model (LLM) at a fraction of the cost typically associated with such systems. This achievement has sent shockwaves across markets, with US tech stocks, particularly in the AI sector, taking a hit as investors reassess the long-held dominance of American companies like OpenAI and Google.
DeepSeek’s rise is emblematic of China’s broader strategy to overcome constraints, maximize innovation, and position itself as a global leader in AI by 2030. This article looks at how DeepSeek has achieved its success, what it reveals about China’s AI ambitions, and the broader implications for the global tech race.
What is DeepSeek, and why does it stand out?
DeepSeek has emerged as a prominent name in China’s AI sector, gaining recognition for its innovative approach and ability to attract top-tier talent. The company, established in 2023 in Hangzhou—a renowned tech hub and home to firms like Alibaba—has recruited exceptional AI researchers from leading universities by offering competitive salaries and opportunities to work on cutting-edge projects. Led by entrepreneur Liang Wenfeng, who also heads its parent firm High-Flyer, DeepSeek has rapidly positioned itself as a key player in the global AI landscape.
Despite operating under constraints, including US restrictions on advanced AI hardware, DeepSeek has demonstrated remarkable efficiency in its development process. Its latest R1 AI model, launched in January 2025, is reported to perform on par with OpenAI’s ChatGPT, showcasing the company’s ability to compete at the highest level. The earlier V3 base model, developed in just two months with a budget of under US$6 million, exemplifies its resource-efficient approach—standing in stark contrast to the billions spent by major US players like OpenAI, Meta, and Anthropic.
This approach challenges traditional assumptions about the costs and infrastructure required to build competitive AI systems, potentially reshaping global perceptions of AI scalability and accessibility. Nvidia, a leading AI chip provider, praised DeepSeek’s innovative use of test-time scaling, a method that enhances performance during model deployment. Such recognition highlights how DeepSeek’s strategy is redefining industry standards, with implications that extend far beyond China.
Where does DeepSeek stand compared to global leaders like OpenAI and Google?
DeepSeek’s AI technology has garnered significant attention for its capabilities, particularly in comparison to established global leaders such as OpenAI and Google. Its earlier model, DeepSeek-V3, demonstrated an impressive ability to handle a range of tasks including answering questions, solving logic problems, and even writing computer programs. These functions were on par with the best AI systems currently available, as shown by standard benchmark tests.
However, just before DeepSeek’s unveiling, OpenAI introduced its own advanced system, OpenAI o3, which some experts believed surpassed DeepSeek-V3 in terms of performance. OpenAI o3 was specifically designed to “reason” through complex problems in areas like math, science, and programming—domains seen as crucial for the future of AI. While DeepSeek had not yet released a comparable reasoning model, many observers noted this gap.
In a surprising move, DeepSeek responded to this challenge by launching its own reasoning model, DeepSeek R1, on January 20, 2025. This model impressed experts across the field, and its release marked a turning point. The AI community, along with global investors, quickly realized the potential significance of DeepSeek’s advancements, triggering widespread concern and market volatility, particularly among US-based investors.
With the debut of DeepSeek R1, the company has solidified its standing as a formidable contender in the global AI race, showcasing its ability to compete with major players like OpenAI and Google—despite operating under significant constraints, including US export restrictions on critical hardware. This development positions DeepSeek as not just a key player in China’s AI sector, but also a rising force on the global stage.
How China is advancing AI innovation despite semiconductor restrictions
China’s progress in AI, despite US semiconductor export restrictions, demonstrates its remarkable ability to adapt and innovate in the face of constraints. The success is driven by three main factors: efficient resource utilization, strategic planning, and a robust AI talent pool. Chinese firms, including start-ups like DeepSeek and tech giants like Tencent, have achieved significant breakthroughs in AI by optimizing the use of less powerful hardware. For example, Tencent’s Hunyuan-Large model outperformed Meta’s Llama 3.1 on multiple benchmarks, showcasing China’s ability to compete on the global stage despite hardware challenges. Similarly, DeepSeek’s new AI model, DeepSeek R1, has garnered attention for matching or even surpassing OpenAI’s ChatGPT o1 in certain benchmarks, but at a fraction of the cost, offering an alternative for researchers and developers with limited resources.
These achievements are largely possible because of advanced software innovations and efficiency techniques that maximize computational output while minimizing hardware requirements. Despite facing restricted access to cutting-edge Nvidia GPUs, Chinese AI labs have been able to produce world-class models, illustrating the importance of algorithmic innovation in overcoming hardware limitations. DeepSeek’s approach, for example, reduced memory usage and sped up calculations without sacrificing accuracy, allowing the company to continue developing high-performing models with limited hardware resources.
A key strategic response to the US export controls has been China’s ability to stockpile Nvidia GPUs prior to the implementation of restrictions. DeepSeek, for example, is believed to have accumulated tens of thousands of these chips, which has ensured continued access to essential resources for training AI models. Some Chinese companies have also resorted to renting GPU access from offshore cloud providers or acquiring hardware through intermediaries to bypass restrictions. These methods have allowed companies to maintain momentum in AI development despite the constraints, highlighting the limitations of the US policy. While the export controls have made it harder for Chinese firms to access cutting-edge hardware, they have not fully stifled China’s AI progress.
China’s vast AI talent pool has been another critical factor in its ability to remain competitive. With a growing number of top AI researchers, surpassing even the US, the country is home to a steady influx of expertise and innovation. Open-source principles are also widely embraced, with companies like Alibaba and start-ups such as DeepSeek and 01.AI contributing to an open AI ecosystem that accelerates development and fosters collaboration. Furthermore, China’s access to extensive datasets and significant government support ensures the continuous flow of talent and resources necessary for pushing AI boundaries.
Despite these advancements, challenges remain. The US export controls aim to limit the scale and efficiency of AI training by reducing access to advanced chips, creating obstacles for Chinese firms developing large-scale models. However, the long-term effectiveness of this strategy is uncertain. As AI development becomes increasingly reliant on high-performance computing, the US may need to reconsider its broad restrictions and shift focus to targeted policies that address specific concerns, such as the development of military AI systems, rather than trying to restrict access to commercial AI technologies.
China’s ability to turn semiconductor restrictions into opportunities for innovation signals its growing resilience and adaptability in the face of geopolitical challenges. The emergence of companies like DeepSeek and its impressive AI models highlights a new phase in China’s AI journey, one marked by increased efficiency, collaboration, and open-source contributions that strengthen its competitive position globally.
Why does China’s AI progress matter on a global scale?
China’s rapid strides in AI are reshaping the global tech landscape, with significant implications for international competition, collaboration, and policy. Despite the challenges posed by US export restrictions on cutting-edge chips, Chinese firms, such as in the case of DeepSeek, are demonstrating that innovation can thrive under resource constraints.
On a global scale, China’s AI advancements are influencing the competitive dynamics between nations and driving new conversations around AI governance. While US companies, including OpenAI, have been focused on enhancing computing power to deliver more sophisticated models, China’s AI ecosystem has taken a different route, prioritizing efficiency and innovation despite hardware limitations. This strategic approach not only narrows the gap between China and the US but also offers a new model of AI development that other nations may look to emulate. The shift in the balance of AI power has broader implications, with countries around the world potentially reassessing their strategies and seeking new opportunities for collaboration with Chinese firms.
Furthermore, China’s AI progress is forcing a rethink of global AI policy. As competition heats up, nations are increasingly focused on regulating AI to manage its ethical and security implications. However, China’s open-source approach, as seen with DeepSeek’s decision to release its best models for free, challenges the paywall-driven model favored by US firms like OpenAI. This raises important questions about access, equity, and innovation in AI. It also reveals the tension between fostering innovation through openness and the need to protect intellectual property and control over cutting-edge technologies.
China’s AI advancements also offer new opportunities for collaboration. With its vast talent pool and commitment to open-source research, China is contributing to a global AI ecosystem where shared knowledge can lead to faster progress. As countries look to harness AI’s potential for economic and technological growth, China’s increasing role as a key player in AI development will shape the future of global innovation and influence AI policy frameworks for years to come.
About Us
China Briefing is one of five regional Asia Briefing publications, supported by Dezan Shira & Associates. For a complimentary subscription to China Briefing’s content products, please click here.
Dezan Shira & Associates assists foreign investors into China and has done so since 1992 through offices in Beijing, Tianjin, Dalian, Qingdao, Shanghai, Hangzhou, Ningbo, Suzhou, Guangzhou, Haikou, Zhongshan, Shenzhen, and Hong Kong. We also have offices in Vietnam, Indonesia, Singapore, United States, Germany, Italy, India, and Dubai (UAE) and partner firms assisting foreign investors in The Philippines, Malaysia, Thailand, Bangladesh, and Australia. For assistance in China, please contact the firm at china@dezshira.com or visit our website at www.dezshira.com.
- Previous Article La trasformazione della supply chain in Cina: innovazioni tecnologiche e prospettive future
- Next Article China’s “National Unified Market” Initiative: 2025 Update