DeepSeek Artificial Intelligence (YP), known in Chinese as Shendu Tansuo (深度探索), meaning “Deep Exploration,” was founded in 2023 in Hangzhou, one of China’s most important IT centers, 1.5 hours away from Shanghai. The company focuses on developing large language models (LLM) and advanced artificial intelligence technologies. The DeepSeek-R1 model developed by DeepSeek has reached a level that can compete with OpenAI’s most powerful models. US President Trump said this was a “Wake Up Call” and announced that they were closely following the process [1].
At what point is DeepSeek-R1 superior?
DeepSeek’s low-cost and efficient YP model can lead to a major paradigm shift in the YP ecosystem. Because DeepSeek changed the understanding that “huge chip power and high cost are required for YP development.” While its competitors are investing billions of dollars in chips and infrastructure, DeepSeek-R1’s development cost was announced as only $6 million.
A token is one of the basic building blocks used as a language unit in natural language processing (NLP) and YP models. A token can be a word, letter, number, punctuation mark, or phrase and is one of the basic building blocks of AI technology. While the processing cost of 1 million tokens for OpenAI o1 was $15, it was only $0.55 for DeepSeek [2], which caused major declines in the shares of US-based AI companies. The world’s largest AI companies, such as Nvidia, lost up to 17% in the stock market [3]. The DeepSeek application reached its peak in the Apple Store, surpassing ChatGPT. Marc Andreessen, a supporter of President Donald Trump and one of the world’s leading technology investors, described DeepSeek as “one of the most surprising and impressive inventions I have ever seen” in an article published in X [4].
It could replace the Nvidia H100.
Since AI requires great expertise, we leave the technical and scientific part of the job to the relevant experts. US companies use Nvidia H100 or higher-end GPUs for more memory bandwidth and higher AI processing power. Chinese AI companies, on the other hand, cannot use H100 or higher GPUs due to the US embargo. Training and running AI models require much data processing and computational power. Higher GPU power and wider bandwidth mean the AI model can work faster, more efficiently, and on a larger scale. However, China-based DeepSeek has broken this mold.
In the AIME 2024 math benchmark test, DeepSeek-R1 achieved a 79.8% success rate, while OpenAI-o1 remained at 79.2%. In addition, despite the limited hardware, it was trained on Nvidia’s H800 GPU clusters instead of Nvidia H100 due to US embargoes. However, DeepSeek announced that this service would be completely open-source, and users can modify its model and create derivative works.
A flower that blooms in China smells nice all over the world.
While Chinese users evaluated it with the saying, “A flower that blooms in China smells nice all over the world,” the founder of DeepSeek said, “China’s AI cannot remain a follower forever… China must have someone who stands at the forefront of technology.”[2]. Can China surpass the US in AI? Of course, it will be quite difficult to answer this question. The US invests almost 10 times more in AI than China[5], and a significant portion of innovative ideas are still US-centered, making the US superior, while China’s investment and technological and scientific fields are progressing quite rapidly. Let’s conclude our article with the meaningful words of the site’s editor, which I benefited greatly from while writing the article and translating it into Chinese.
“From the atomic bomb and a satellite project to the Beidou Navigation System, from wind tunnel technology to quantum communication… At what point have we not succeeded by advancing step by step? At what point have we not fallen and stood up again many times? Accepting this difference and moving forward with hard work confirms the saying, ‘No matter how long the road is, if you keep walking, you will reach it’ and reveals the understanding that ‘When success is achieved, it is shared with everyone.'”
Resources
1-) https://www.theguardian.com/technology/2025/jan/28/donald-trump-china-deepseek-ai-chatbot-shares
[2] https://baijiahao.baidu.com/s?id=1822444794434949501&wfr=spider&for=pc
[3]https://www.reuters.com/technology/tech-stock-selloff-deepens-deepseek-triggers-ai-rethink-2025-01-28/
[4] https://edition.cnn.com/2025/01/27/tech/deepseek-stocks-ai-china/index.html
[5] https://www.163.com/dy/article/JN3CRL4305199NPP.html
Africa Ankara Asia Belt & Road Belt & Road BRICS Carbon Emission China CIIE CSC Culture Declaration Development Economy education elder EU Europe Global Economics HTS Kültür MEB Middle asia movie Multipolarity Neoliberalism Russia scholarship Shanghai Sino Turkish Sino Turkish Studies Sino Turkish Studies Syria Taiwan trump Turkey Turkiye Two Sessions Türkiye USA Uyghur Vatan Party Xi Jinping Xinjiang Çin
