DeepSeek: The Chinese AI Company That Took The World By Surprise

The logo of DeepSeek, with a prominent blue whale.

In January 2025, the global tech landscape was shaken by the release of a groundbreaking AI model whose performance benchmarks rivaled—and in some cases surpassed—OpenAI’s AI models (like ChatGPT). In an industry dominated by Western innovation, this development was nearly unheard of. Even more surprising was the source, as DeepSeek AI was not a model developed in Silicon Valley. Rather, it was created in China, a country viewed by the West as a geopolitical adversary. 

What is DeepSeek?

Based out of Hangzhou, China, DeepSeek is an AI development firm, and it was founded in May 2023 by Liang Wenfeng, a graduate of the prestigious Zhejiang University. Moreover, Liang Wenfeng established High-Flyer, a China-based quantitative hedge fund, which is a type of investment fund that relies on mathematical models and statistical analysis to make investment decisions. Currently, DeepSeek operates an independent AI research lab under High-Flyer.

DeepSeek’s first model was released in November 2023, but the company really only got famous after the debut of the DeepSeek-R1 reasoning model in January 2025. DeepSeek-R1 is notable for its high-achieving performance levels, especially while using fewer resources to create and operate. For example, OpenAI’s GPT-4 model required hundreds of millions of dollars to develop, whereas DeepSeek’s R1 model required less than 6 million dollars.  Furthermore, it’s much more cost-effective to use DeepSeek R1, as R1’s input tokens cost only $0.55 per 1 million tokens, compared to OpenAI’s GPT-4 model’s cost of $5 per 1 million input tokens. That means, on average, DeepSeek’s R1 model is over nine times cheaper to use than OpenAI’s GPT-4 model!

The reason for why DeepSeek is so effective at such a low cost is its different approach to training its R1 models. First, DeepSeek emphasizes Mixture-of-Experts (MOE) Architecture, meaning that whenever someone inputs a question into the AI model, DeepSeek activates only a few specialized “experts” per input instead of activating all neural network components at once. Essentially, DeepSeek utilizes only the most important and necessary parts of its processing capabilities to answer the input. Second, DeepSeek changes its AI models based on user feedback and real-world use cases. For example, if users find a problem while using one of DeepSeek’s AI models, DeepSeek will fine-tune it until that mistake is fixed. 

Global Reception and Impact

When DeepSeek was first released into the Western market, it heavily disrupted the international AI market. DeepSeek’s emergence had a significant, albeit short–term, impact on the stock market, as its low-cost and efficient AI model triggered a massive tech sector sell-off, which is when a pool of investors decides to sell a large quantity of securities (eg. stocks) in a very short period of time, typically leading to a sharp decline in the value of the securities. In this case, the sell-off caused Nvidia’s stock to plummet 17%, and it erased nearly $1 trillion in market capitalization. 

Not only were there financial consequences, but there were also many concerns, especially regarding privacy, in the Western world. DeepSeek’s effectiveness, while still being cost-effective, faced criticism and scrutiny from American politicians. 

Some politicians attempted to downplay the significance of DeepSeek by claiming that DeepSeek relied on a covert supply of advanced graphics processing units (GPUs) from abroad, while others claimed that DeepSeek is a national security threat and attributed DeepSeek’s success to China stealing U.S. technology. Unsurprisingly, several members of the House of Representatives, such as Darin LaHood, an Illinois Republican, and Josh Gottheimer, a New Jersey Democrat, introduced a bill that would ban DeepSeek’s applications and AI models from government-owned devices. 

This negative response wasn’t just from the United States; it also came from countries like Australia, where Strategic Analysis Australia, a national security consulting firm, demanded DeepSeek be banned in the country. They warned that it “could give the People’s Liberation Army a significant combat edge, just as geopolitical tension will see the probability of confrontation and even conflict rise over the next four to five years.” Many other Western countries and allies are cautious and warning against the future usage and development of DeepSeek’s AI models. 

Furthermore, DeepSeek’s entrance into the global AI market has called into question China’s censorship policies. Deepseek’s AI models follow China’s censorship and restriction policies, which prevent the generation of responses related to politically sensitive stuff like the Tiananmen Square Incident and Taiwan’s Sovereignty. When faced with prompts involving such topics, the AI model will either divert the conversation by saying something like “Sorry, that’s beyond my current scope. Let’s talk about something else” or avoid responding to the input altogether. 

Conclusion

Despite some bad reception, DeepSeek was revered by the majority for its innovative and financially cost-effective development and usage. According to surveys with populations in Western countries, 57% of respondents have experimented with DeepSeek-R1 and are interested in DeepSeek’s AI models. 

DeepSeek’s rise in fame represents a new chapter in the global tech landscape, where, despite backlash from international companies and its limitations through censorship, it has captured global attention because of its cutting-edge technology and the resulting discomfort in Western governments. Whether welcomed or feared, DeepSeek has been defining what it means to develop and create artificial intelligence models.