A Chinese AI company has panicked Wall Street tech investors with its cost-efficient AI model rivaling giants such as OpenAI and Google.
DeepSeek developed the model with around $6 million, a small fraction of the cost that other companies incurred training their models, and released it on Jan. 20. OpenAI spent billions developing its ChatGPT model in 2024.
Immediate effects
The stock market reacted strongly to the company’s advance on Jan. 20, with the S&P 500 sinking nearly 2% and tech companies facing the brunt of the sell-off. Nvidia, which develops a large number of the United States’s graphic processing units that power AI, had its stock drop about 17% as of 2 p.m. Tuesday.
“DeepSeek AI’s entry into the market offers a disruptive combination of high-level capabilities at a fraction of the cost of industry giants like OpenAI,” Corpora.ai CEO Mel Morris told the Washington Examiner. “This affordability opens the door for smaller companies and startups to leverage advanced AI technology that was previously inaccessible. Additionally, the open-source model empowers developers to fine-tune and experiment with the system, fostering greater flexibility and innovation.”
Morris said he believes DeepSeek will push the AI industry to be more efficient.
“The emergence of DeepSeek AI is poised to challenge the current competitive landscape, pushing major players like OpenAI to adapt quickly,” he added. “The idea that competition drives innovation is particularly relevant here, as DeepSeek’s presence is likely to spur faster advancements in AI technology, leading to more efficient and accessible solutions to meet the growing demand.”
What has DeepSeek done?
The model uses much fewer computer chips than the top AI companies in the U.S. thanks to an innovative approach to maximize computing power. The lack of access to computer chips is due to the broad restrictions that former President Joe Biden’s administration imposed on the sales of the materials to China amid a tech rivalry with the U.S.
That chip ban spurred innovation from the company, which did not struggle with obtaining the money it might have needed to buy more.
“The problem we are facing has never been funding but the export control on advanced chips,” DeepSeek founder Liang Wenfeng told Chinese tech publication 36Kr last year.
The efficiencies they accomplished are not new, an expert on AI said, but combining them is impressive.
“They optimized their model architecture using a battery of engineering tricks — custom communication schemes between chips, reducing the size of fields to save memory, and innovative use of the mix-of-models approach,” Wendy Chang, a software engineer at the Mercator Institute for China Studies, told Wired. “Many of these approaches aren’t new ideas, but combining them successfully to produce a cutting-edge model is a remarkable feat.”
Liang also attributes his company’s success to the youth of its developers. He says they mostly hire people who have recently graduated from college and can “devote themselves completely to a mission without utilitarian considerations.”
Marina Zhang, an AI expert and associate professor at the University of Technology Sydney, said patriotism also plays a factor.
“This younger generation also embodies a sense of patriotism, particularly as they navigate U.S. restrictions and choke points in critical hardware and software technologies,” Zhang said. “Their determination to overcome these barriers reflects not only personal ambition but also a broader commitment to advancing China’s position as a global innovation leader.”
AI bias
China’s bias is shown in the model. DeepSeek’s AI refuses to answer questions about the 1989 Tiananmen Square massacre and will not criticize Chinese President Xi Jinping. When asked about either topic, the model will decline to answer. “Sorry, that’s beyond my current scope. Let’s talk about something else,” it says when asked about the protests.
OpenAI’s ChatGPT model is known for not answering some political questions in the past, including most questions surrounding President Donald Trump. It will now answer political questions pertaining to Trump or former Vice President Kamala Harris, but when asked during the campaign season to endorse a 2024 presidential candidate, it declined to do so. “As a neutral AI, I can’t make a personal judgment or endorse one over the other,” it said.
The model is just as capable as American AI models and is also especially good at math and coding.
The company provides a formidable opponent for U.S. AI companies, which have operated with the comfortability of numerous computer chips necessary to develop AI models. DeepSeek’s efficiency could turn their thought process upside down, though it has drawn early questions.
CLICK HERE TO READ MORE FROM THE WASHINGTON EXAMINER
The company claimed to only have used 10,000 Nvidia A100s in 2024, but Scale.ai founder Alexandr Wang accused it of holding 50,000 Nvidia H100s, a significantly larger stockpile that could draw into question the company’s efficiency. The H100 costs around $25,000 per graphics processing unit, which means if Wang’s estimate is correct, DeepSeek holds $1.25 billion worth of GPUs.
Wang said the company cannot talk about the chips because of the export controls the U.S. has in place with China.