%202.png)
By Kimberley Sun
The Lawrenceville School, NJ
Following OpenAI’s launch of ChatGPT, the AI chatbot industry has grown exponentially, projected to reach $27.2 billion by 2030. The market is based on many advantages of AI, including its accessibility to people around the world, personalization of feedback that human resources are unable to mimic, and the speed of its responses. AI models have grown quickly from simplistic rule-based systems to intelligent learning models, pulling on tools such as machine learning and neural language processing to fine-tune their performances.
Deepseek’s Chatbot
On Monday, Nvidia shares plummeted 17%, wiping out $600 billion in market value and setting a record for the largest single-day loss ever suffered by a company in stock market history. The cause of this? A small Chinese company based in Hangzhou — Deepseek.
In December, Deepseek released a chatbot that rivaled the performance capacities of current tech giants such as OpenAI and Google. This chatbot supposedly took only two months and less than $6 million to build, demonstrating a remarkable leap in AI development efficiency. Furthermore, the results of American benchmark tests showed that it not only matched competitive chatbots on the market but, in some cases, such as AIME 2024, even surpassed them. Additionally, Deepseek has open-sourced its AI system, allowing groups other than tech giants to understand and modify the technology.
Benchmark performance comparison of AI chatbot models
Aside from its impressive performance, the Deepseek model caused a significant buzz in the tech industry mainly for two reasons: Deepseek uses less money and fewer chips.
Compared to other chatbot systems that used tens of hundreds of millions in raw computing power, Deepseek only used $6 million in its development, nearly ten times less than what Meta’s chatbot needed. Due to this significantly lower development cost, Deepseek also had a much reduced price for its users, almost 27 times less than that of ChatGPT’s o1 model. This raised a concern for other companies in terms of their price competitiveness, and many companies also reduced their prices. However, since their development cost was dramatically higher than that of Deepseek, many companies’ revenues also dropped accordingly, some even to the point of loss.
Arguably, however, Deepseek’s greatest accomplishment was not its low model cost, but its chip efficiency. Incredibly, the Deepseek chatbot only used 2000 Nvidia GPUs. For context, ChatGPT used 10000, and many other major tech companies used more. US chip sanctions on China had sought to limit their ability to develop advanced AI technology but instead forced Chinese engineers to be more efficient and innovative with their designs. As observed, Deepseek doesn’t aim to show detailed explanations and thought processes compared to other larger models. Instead, it focuses on giving accurate answers, which greatly reducing its memory and computing time. Specifically, Deepseek’s new reasoning model, r1, displayed a chain of thought process. In other words, it was able to self-reflect and reevaluate its previous logic to refine its future answer. This ability allows it to solve more complex problems with a greater accuracy than other models.
Image of a computer server room
Why is this Deepseek’s greatest accomplishment? Lower memory usage means higher energy efficiency, one of the most imminent issues for advancing AI technology today. As AI usage increases, so does its consumption of energy. The energy consumption of data centers is projected to reach 1000 Terawatts by 2026, double that of 2022 and the equivalent of Japan’s total energy consumption. While steps have been taken to ensure that AI energy consumption does not break the roof, the data that is needed for those regulations to be truly impactful are very much missing. Deepseek’s higher energy efficiency is a step to combat AI’s contribution to climate change. Since it is open-source, other tech developers can also learn from its system to enhance their own AI models.
Deepseek is currently not commercializing its models, but focusing on research to “promote the development of the entire ecosystem”, as Deepseek founder Liang Wenfeng said, through sharing the technology with the entire industry. While the Deepseek models may give rise to competition within the industry for monetization, it is undoubtedly a huge milestone in the development of AI chatbot models, signifying a shift into a more sustainable and efficient AI future.
---- Works Cited
Berreby, David. “As Use of A.I. Soars, so Does the Energy and Water It Requires.” Yale E360, 6 Feb. 2024,
Chen, Caiwei. “How a Top Chinese AI Model Overcame US Sanctions.” MIT Technology Review, 24 Jan. 2025,
Kasanmascheff, Markus. “How DeepSeek R1 Surpasses ChatGPT O1 under Sanctions, Redefining AI Efficiency Using
Only 2,048 GPUs - NewsBreak.” NewsBreak, Winbuzzer.com, 25 Jan. 2025, www.newsbreak.com/winbuzzer-com-302470011/3777104679841-how-deepseek-r1-surpasses-chatgpt-o1-under-sanctions-redefining-ai-efficiency-using-only-2-048-gpus.
Metz, Cade, and Meaghan Tobin. “How Chinese A.I. Start-up DeepSeek Is Competing with OpenAI and Google.” The New
York Times, 23 Jan. 2025, www.nytimes.com/2025/01/23/technology/deepseek-china-ai-chips.html.
Sherman, Ben. “AI Chatbots: Understanding the Benefits and Limitations.” Synoptek, 17 Nov. 2023,
synoptek.com/insights/it-blogs/data-insights/ai-chatbots-understanding-the-benefits-and-limitations/
Subin, S. (2025, January 27). Nvidia sheds almost $600 billion in market cap, biggest one-day loss in U.S. history.
Retrieved January 25, 2025, from CNBC website: https://www.cnbc.com/2025/01/27/nvidia-sheds-almost-600-billion-in-market-cap-biggest-drop-ever.html
Turner, Ben. “Chinese Researchers Just Built an Open-Source Rival to ChatGPT in 2 Months. Silicon Valley Is Freaked
Out.” Livescience.com, Live Science, 24 Jan. 2025, www.livescience.com/technology/artificial-intelligence/china-releases-a-cheap-open-rival-to-chatgpt-thrilling-some-scientists-and-panicking-silicon-valley.
“揭秘DeepSeek:一个更极致的中国技术理想主义故事 ｜36氪独家-36氪.” 36kr.com, 27 Jan. 2025,
Comments