Despite the public release of its latest model being barely two months prior, Deepseek has become a household name for AI researchers, students, and corporate executives alike. Millions of people have now used the most popular and advanced Chinese AI model to date, and it appears to have neatly found its niche within the current AI ecosystem. However, the rise of Deepseek means much more than just offering users another alternative to ChatGPT on the market; it represents a possible shift in the current state of American AI dominance and showcases the results of longstanding Chinese policy on tech investment.

To understand what impacts Deepseek are likely to have, a brief dive into the history of the company behind it and its history is necessary. “Deepseek” as colloquially used more often than not refers to the Deepseek-R1 Large Language Model (LLM), the most recent LLM release by Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Company. The company was formally founded in 2023 but had been operating long before then as a group within High-Flyer, a Chinese hedge fund that used AI for its investments.^[1] By 2021, High-Flyer had already stocked up on thousands of Nvidia’s AI chips for developers working on earlier iterations of its LLMs. However, it received little attention in the west until December of last year, when it released its Deepseek-V3 LLM model, notable for matching the capabilities of the most advanced publicly available LLM models from OpenAI and Google.^[2] Just weeks later, Deepseek proceeded to release both a refined version of the model, Deepseek-R1, as well as an associated chatbot available both on its website and in the form of an app that quickly rose to the top of app store download charts.^[3]

While impressive, the development of a rival to current AI models on the market was hardly groundbreaking news. What truly shocked the tech world about Deepseek-V3 was the fact that Deepseek-V3 cost 6 million USD to train and build, compared to the staggering estimate of 100 million USD spent on OpenAI’s O1 model.^[4] When it comes to AI models, training is everything, requiring massive amounts of data to be fed through computers that process it to teach models how to produce coherent outputs. Common consensus had always been that computing power was the only way to see improvements in AI, with the US previously banning the export of top-shelf machine learning chips produced by Nvidia to China as an attempt to retain its dominance in tech. Deepseek, however, appears to have completely upended this understanding of machine learning, claiming in a research paper that its new model only required 1/8^th of the chips that current AI powerhouses trained their models on.^[5] Despite this massive resource disparity, testing found that Deepseek’s R-1 model had comparable performance to OpenAI’s current flagship o1 model, even managing to beat o1 in mathematics and coding.^[6]

Both the rise of Deepseek, as well as the manner in which it occurred, poses massive implications for the future of global AI development. Deepseek’s results and the unexpected publicity that they’ve faced in the west marks the first major challenge to America’s previously unrivaled status as the center of AI research, and the Chinese government quickly jumped on this opportunity. Chinese news networks have been quick to praise Deepseek as an example of the technological innovation that the Chinese government had attempted to foster, and over 20 state-owned enterprises have begun using Deepseek in their day-to-day operations after the central government called for greater implementation of AI to improve efficiency. ^[7] It is almost a certainty, then, that elites in the CCP look favourably upon the chances of Deepseek joining the ranks of Chinese tech giants like Huawei or Alibaba. More importantly, however, Deepseek’s optimization of machine learning has shattered current US strategies of AI dominance and containing Chinese technological development. Increased calculation efficiency means that computing power is no longer the be all and end all of AI innovation, potentially weakening the current America containment strategy of heavily limited chip exports to China. This also makes deepseek a far more competitive candidate for use in the third world compared to OpenAI’s models; when researchers can run similarly taxing experiments on Deepseek-R1 for 1/30^th of the price of doing so on OpenAI’s o1 model, it can be expected that many economically disadvantaged countries are likely to see a boom in AI adoption that did not come with the development of previous AI models.^[8]

Currently, a number of governments have put forward bills to ban the usage of Deepseek. Australia, Italy, and Taiwan have all moved to limit its presence in their nations,^[9] with the US recently tabling legislation to ban it on government devices as well.^[10] However, a national ban on Deepseek fails to target it where it could potentially hurt Western dominance of AI the most: in countries abroad, where the cheap and efficient performance of Deepseek’s models pose a powerful temptation to governments and researchers alike. No matter what the response to it is, however, one thing is clear: Deepseek has upended the current understanding of AI, and marks the beginning of a new era for technological competition.

---

Jack is a current third-year student at the University of Toronto with majoring in International Relations and Peace, Conflict, and Justice, and a minor in Contemporary Asian Studies. His research interests involve the rise of China and its international implications, with past research experience as a member of the G20 research group and a volunteer for the Canada-China Business Council.

---

Bibliography

“Which Countries Have Banned DeepSeek and Why?.” Al Jazeera. Al Jazeera. Accessed February 6, 2025. https://www.aljazeera.com/news/2025/2/6/which-countries-have-banned-deepseek-and-why.

“DeepSeek.” Deepseek. Deepseek. Accessed February 6, 2025. https://www.deepseek.com/.

Gibney, Elizabeth. “China’s Cheap, Open AI Model DeepSeek Thrills Scientists.” Nature 638, no. 8049 (January 23, 2025): 13–14. https://doi.org/10.1038/d41586-025-00229-6.

Jian, Yong. “DeepSeek Is Now the Brain of Chinese State-Owned Firms.” Asia Times. Asia Times February 28, 2025. http://asiatimes.com/2025/02/deepseek-is-now-the-brain-of-chinese-state-owned-firms/.

Leigh, Harri. “Jon Husted Introduces First Bill to Ban Chinese DeepSeek from Government Devices.” Spectrum News. Spectrum News. February 24, 2025. https://spectrumnews1.com/oh/columbus/news/2025/02/28/husted-deepseek-ban-ai.

Metz, Cade. “What to Know About DeepSeek and How It Is Upending A.I.” The New York Times. The New York Times. January 27, 2025. https://www.nytimes.com/2025/01/27/technology/what-is-deepseek-china-ai.html.

1. Cade Metz, “What to Know About DeepSeek and How It Is Upending A.I.,” The New York Times, January 27, 2025, https://www.nytimes.com/2025/01/27/technology/what-is-deepseek-china-ai.html. ↑

2. “DeepSeek,” accessed February 6, 2025, https://www.deepseek.com/. ↑

3. “Which Countries Have Banned DeepSeek and Why?,” Al Jazeera, accessed February 6, 2025, https://www.aljazeera.com/news/2025/2/6/which-countries-have-banned-deepseek-and-why. ↑

4. Metz, “What to Know About DeepSeek and How It Is Upending A.I.” ↑

5. Metz. ↑

6. Elizabeth Gibney, “China’s Cheap, Open AI Model DeepSeek Thrills Scientists,” Nature 638, no. 8049 (January 23, 2025): 13–14, https://doi.org/10.1038/d41586-025-00229-6. ↑

7. Yong Jian, “DeepSeek Is Now the Brain of Chinese State-Owned Firms,” Asia Times, February 28, 2025, http://asiatimes.com/2025/02/deepseek-is-now-the-brain-of-chinese-state-owned-firms/. ↑

8. Gibney, “China’s Cheap, Open AI Model DeepSeek Thrills Scientists.” ↑

9. “Which Countries Have Banned DeepSeek and Why?” ↑

10. Harri Leigh, “Jon Husted Introduces First Bill to Ban Chinese DeepSeek from Government Devices,” Spectrum News, February 24, 2025, https://spectrumnews1.com/oh/columbus/news/2025/02/28/husted-deepseek-ban-ai. ↑