An artificial intelligence (AI) company from China that operated discreetly until now began to gain space in the global market. It is DeepSeek, a company that already worries giants such as OpenAI and Nvidia and heated up the war between the United States and the Asian country in this busy sector.
The DeepSeek language model was presented last Monday (20) and would perform even better than the current ChatGPT engine according to some tests. In addition, the chatbot based on it suddenly became popular, to the point of becoming the most downloaded free app from the App Store in recent days.
The first numbers had an immediate impact on the speculative market. Nvidia’s shares, until then one of the companies with the highest market value in the world, fell 16% in a single day as a reflection of the importance that the company has today in supplying GPUs for AI systems of other brands. Nasdaq itself, the stock exchange most linked to technology companies, had a 3% drop due to investor concerns about a possible overtaking of the Chinese in the sector.
DeepSeek’s servers are overloaded with so many accesses, and registrations have been limited indefinitely until the situation is resolved. Still, with a Google account, you can test the free version of the chatbot through this link. The tool can also be downloaded for Android and iOS.
The origin of DeepSeek
DeepSeek was founded in 2023 by businessman and engineer Liang Wenfeng. Nothing media, he said in the few interviews he has given so far, that he did not expect the sudden success of the brand or that the economic factor was so relevant to the market.
The current language model, called DeepSeek-R1, is open source and would have been developed with a much smaller budget than the competition. The numbers are still considered suspicious, but the training alone would have cost less than $6 million, while Anthropic, from the chatbot Claude, cited expenses that go to almost $1 billion on the same task. Even higher values are estimated as investments in the area by names like Meta, OpenAI, and Google.
The cost of AI operation and licensing is lower, attracting companies interested in partnerships. This is the main reason for the drop in US company stocks. Shareholders are concerned about overspending, It can be offered at much lower prices.
How does she manage to compete with the US?
The figures released so far by DeepSeek indicate that it can do better than OpenAI’s GPT-4o while being more efficient and faster than its American rival.
Interestingly, the company only achieved its current performance because of the United States trade sanctions. The country has for months been barring US companies or allies from exporting semiconductors and other AI materials to companies in China, which led DeepSeek to seek internal solutions.
With Nvidia’s own modern GPUs banned, such as the H100, DeepSeek sought regional solutions and smaller companies, capable of achieving satisfactory results already knowing these limitations. The CEO himself was able to buy a large amount of Nvidia’s own cards before the suspension, which helped energize part of the initial project.
Even Microsoft CEO Satya Nadella is amazed by DeepSeek in a speech at the World Economic Forum in Switzerland. According to him, it is impressive how much AI has managed to do in terms of efficiency and data processing, even more so as an open-source model.
DeepSeek, for now, still has little transparency regarding training methods and if it used copyrighted materials, such as competitors, and energy issues, such as consumption for each requested prompt. More technical details should be provided by the brand soon.
In addition, the excitement with the Chinese company may draw the attention of the US government: it is not difficult to imagine that it is the target of US sanctions. As well as equipment, components, and social networks in the region, which could even prohibit its use in the country.
