They argue that DeepSeek demonstrates China’s increasing capability to develop cutting-edge AI models independently of Western technology.
On January 27, 2025, just a day after China celebrated the Spring Festival, Chinese AI startup DeepSeek delivered a significant blow to the U.S. tech sector with its “low-cost + open-source” (低成本+开源) model. In a single day, U.S. tech companies lost nearly $1 trillion in market value, with NVIDIA alone accounting for nearly $600 billion of the decline.
This massive wealth erosion reflects not only shaken confidence in U.S. dominance in artificial intelligence but also DeepSeek’s emergence as a formidable competitor. The startup now stands alongside China’s tech giants—Huawei, Baidu, Alibaba, Tencent, and ByteDance—as yet another major disruptor in the AI landscape.
Focused on large language models (LLMs), AI-powered search, and foundational AI infrastructure, DeepSeek has challenged Western AI dominance and is poised to play a crucial role in shaping China’s future tech landscape. An article by All Weather TMT, a Chinese website dedicated to science and technology, described DeepSeek’s success as a “technological breakthrough” (技术突破) and heralded the arrival of the “DeepSeek Moment.”
This breakthrough has compelled leading Chinese and global tech giants—including Huawei, Tencent, Baidu, ByteDance, Alibaba, Microsoft, Amazon, NVIDIA, and Intel—to integrate DeepSeek-R1/V3 inference services into their cloud platforms. Regarding cloud platforms, the article asserts that “whoever secures the most developers will have the upper hand in the next AI boom.”
China’s official news agency, Xinhua, argued that the shockwaves caused by DeepSeek stem from its significantly lower costs and efficiency. Compared to ChatGPT-4o, DeepSeek operates at just 1/18th of the training cost, with a team one-tenth the size, while delivering comparable model performance. According to Xinhua, the API pricing for DeepSeek-R1 is: ¥1 per million input tokens (cache hit), ¥4 per million input tokens (cache miss), and ¥16 per million output tokens. In contrast, ChatGPT charges ¥55 per million input tokens, ¥110 per million input tokens, and ¥438 per million output tokens, respectively.
This pricing structure allows DeepSeek to disrupt the Western “cash-burning business model” (烧钱模式) with its “optimum training costs and computational efficiency”—a strategy encapsulated by the phrase “achieve much with little effort” (四两拨千斤). Furthermore, Xinhua highlighted that DeepSeek-V3 was trained with just $5.576 million on a compute-limited NVIDIA H800 GPU cluster, whereas OpenAI’s GPT-4o required a $100 million investment in high-performance NVIDIA H100 GPUs.
On February 8, The Paper highlighted DeepSeek’s performance compared to ChatGPT-4o across various benchmarks. According to the report, DeepSeek-R1 scored 79.8% on AIME 2024 (American Invitational Mathematics Examination), surpassing OpenAI’s o1-1217 score of 79.2%. On MATH-500, a benchmark for math competition problems of varying difficulty, DeepSeek-R1 achieved an impressive 97.3%, performing on par with OpenAI o1-1217.
For coding-related tasks, DeepSeek-R1 demonstrated expert-level proficiency, earning an Elo rating of 2029 on Codeforces—outperforming 96.3% of human participants. In engineering-related tasks, DeepSeek-R1 slightly outperformed OpenAI o1-1217. Unsurprisingly, The Paper noted that within just 18 days of its launch, DeepSeek had been downloaded 16 million times—nearly double the 9 million downloads that OpenAI’s ChatGPT achieved during its initial release.
Li Dian, a researcher at the Institute of China Studies, Fudan University, asserts that DeepSeek “is not challenging the rules; it’s rewriting the game” while also reshaping the geopolitical AI landscape. Numerous articles in Chinese media have suggested that DeepSeek’s success directly challenges the relevance of capital-intensive AI projects—such as the $500 billion StarGate initiative and SoftBank’s $25 billion investment in OpenAI, while also calling into question the effectiveness of high-tech sanctions imposed on China.
They argue that DeepSeek demonstrates China’s increasing capability to develop cutting-edge AI models independently of Western technology. An article in the ST Daily supports this view, stating that by adopting an unconventional approach, DeepSeek has carved out a niche, proving that Chinese enterprises are no longer mere followers or imitators. True innovators, the article asserts, cannot be constrained by technological blockades. However, these claims have been challenged by Western media. The Financial Times, in an article, reported that “OpenAI says it has evidence China’s DeepSeek used its model to train a competitor.”
Professor of economics Xu Chenggag argues that DeepSeek’s success cannot be regarded as a breakthrough, as it has not introduced “new scaling laws” in AI, unlike OpenAI. However, this argument is disputed by scholars such as Jia Kai, Jiang Yuhao, Dai Mingjie, and Zhang Xian, all fellows at the Guangdong New Quality Productive Forces Policy Research Center. They contend that high investment no longer guarantees high performance, leading more technology pioneers, corporate leaders, and prominent investors to openly question the validity of the Scaling Law.
Discussing DeepSeek’s “unprecedented implications for military AI and national security,” Xu Bingjun, a Research Fellow at Xinhua News Agency’s Liaowang Think Tank, highlights six key opportunities and transformations in military AI: accelerating the development of military intelligence, enhancing strategic deterrence, improving decision-making, boosting combat effectiveness, strengthening autonomous combat capabilities, and optimizing military training.
Xu believes DeepSeek’s unique “chain of thought” reasoning architecture and algorithmic innovations will be applied in intelligent combat systems, weapon design and simulation, cybersecurity and defence, and civil-military fusion. He predicts that DeepSeek could “disrupt existing military balances” and enable other nations to enhance their AI-driven military capabilities. Xu concludes that it may even “mark the end of the AI-based computational arms race” (算力军备竞赛).
While many analysts worldwide have acknowledged DeepSeek’s low-cost, high-performance, and open-source model, concerns over data security and cybersecurity have led several governments—including those of South Korea, Italy, Australia, India, the U.S., and Japan—as well as enterprises to reportedly restrict or ban access to DeepSeek. Former U.S. President Donald Trump stated that DeepSeek’s rise “should be a wake-up call” for America’s tech companies.
On February 6, China’s Foreign Ministry spokesperson Guo Jiakun addressed these reports, emphasizing that “the Chinese government places great importance on data privacy and security.” He further stated that China firmly opposes the politicization of economic, trade, and technological issues under the pretext of national security and reaffirmed the country’s commitment to resolutely safeguarding the legitimate rights and interests of Chinese enterprises.
On February 7, Loongson Technology Corporation Ltd. announced that devices powered by Loongson 3 series CPUs had successfully deployed and run the DeepSeek R1 7B model locally. This breakthrough eliminates reliance on cloud servers, mitigating risks associated with network instability or server overload. However, concerns persist regarding DeepSeek’s ties to the Chinese Party-State, a concern often raised about other Chinese enterprises.
Some argue that due to ideological constraints, DeepSeek avoids addressing sensitive topics such as Tiananmen and Tibet. However, the model does not appear to exhibit such inhibitions—it explicitly refers to the “Tiananmen Square incident, also known as the Tiananmen Square Massacre,” and provides a standard explanation of the “Tank Man.” Even if DeepSeek is perceived as state-driven rather than market-driven, it is likely to gain traction in non-Chinese-speaking countries—unlike WeChat, which remains primarily limited to Chinese-speaking users.
DeepSeek’s rise underscores the broader evolution of China’s innovation ecosystem, drawing parallels to Japan’s technological advancements in the 1980s and 1990s. Key factors such as open-source databases, government subsidies, and the seamless integration of technology across domestic industrial supply chains have created a fertile environment for high-tech innovation, with AI being no exception. Therefore, DeepSeek’s achievements should not be seen in isolation. It has undeniably intensified pressure on long-established international tech giants and may drive AI enterprises to reconsider and redefine industry standards as AI continues to proliferate in the future.
* B.R. Deepak is Professor, Center of Chinese and Southeast Asian Studies, Jawaharlal Nehru University, New Delhi.