Neway weapons race

The Deepseek Chinese Chat Botto bypasses ChatGPT in popularity and collapses world markets. Are we close to the Revolution of the II Industry?

In January, a new favorite appeared on the chat bots market-the Chinese Deepseek neural network. Her mobile application has become the most popular in the American App Store, overtaking ChatGPT. The creators of DeepSeek claim that they spent less than six million dollars (many times less than competitors) on computing capacities for training in a model, and it surpasses all analogues according to the test results. It was previously believed that only leading American companies with billions of dollars can create the best tools using artificial intelligence.

On January 27, the US stock market responded to the emergence of a Chinese neural network by a drop in shares by 3.1% - more than a trillion of dollars. And NVIDIA securities fell into more than 17%, which amounted to 593 billion dollars-this is the largest fall in the history of Wall Street. According to the host of the BBC, Mark Tseslak, the appearance of DeepSeek means only one thing: the states can no longer be called the main neuruslate power. New Europe says what you need to know about the sensational Deepseek: who created this neural network and what its features have changed, what has changed in the AI ​​market and whether the United States has a comparable analogue.

What is the remarkable Deepseek

Until recently, this neural network was known only in narrow circles of enthusiasts. The company with the same name in 2023 was founded by a 40-year-old Chinese billionaire Liang Wanfen, the owner of a large hedge fund High Flyer . In 2021, Wenfenus offered business partners the idea of ​​Ai-Startap and bought thousands of NVIDIA video cards to teach his own neural network. But he was responded with a refusal: many believed in Boom AI, but not in the success of a small company. It was believed that large Chinese corporations like ByTedance, which owns Tiktok, will first create a breakthrough neural network.

But in 2022, the United States banned the Taiwanese corporations NVIDIA and AMD to sell powerful video cards to China. American officials were afraid that with their help the PRC could create advanced military technologies, including in the field of artificial intelligence. Due to the sanctions, China did not receive advanced video cards for hundreds of millions of dollars and the next year began to lag behind the United States in a neural network race.

Under these conditions, Venfan's startup began to look much more tempting. The entrepreneur managed to buy thousands of NVIDIA video cards and in May 2023 gathered a team that began work on the Deepseek neural network. His startup did not have access to the latest technologies, so it cost relatively cheap and obsolete NVIDIA H800, which were inferior in performance by the latest H100.

In November of the same year, the company released the first version of Deepseek, and in May 2024-the second. But the company became truly known only in January 2025, when it published in the public domain a “reasoning” model of Deepseek R1 with the possibility of double-checking, conduct a deep analysis and compile logical chains. At the same time, Deepseek emphasized that

It took only two months and $ 5.6 million to study R1, although American companies spend billions on their chat bots. The model surpassed the advanced ChatGPT-O1 by solving complex mathematical problems in three of the six tests.

Deepseek is available for free all over the world, including in Russia without VPN, on browsers, iOS and Android. The limit for use is 50 messages per day for the base model V3 and 10 for the “reasoning” R1. The neural network has open source code: any person or company, if desired, can use Deepseek achievements to create their own model. The price of the API is 95% lower than that of ChatGPT.

Deepseek has its own restrictions. Due to the influx of users, the neural network can significantly delay the answers, periodically the site falls due to the load. Already, the company has limited registration for users with Chinese phone numbers, but you can enter the Google or Apple account.

Also, the chatbot has government censorship. So, DeepSeek refuses to answer what happened on June 4, 1989 on Tyankanman Square and Hong Kong in 2019. In the same way, the neural network does not answer questions with a negative subtext about the chairman of China Xi Jinping and, in general, does not allow criticizing the China Communist Party.

Despite the disadvantages, Deepseek became the most discussed by the neural network in early 2025. Thanks to this, Lyan Wenfen was invited to a meeting with the Prime Minister of the State Council of China Lee Cyan. Tsyan called the technical innovations the driver of economic growth and discussed the work of the Chinese government with the head of Deepseek.

How the Chinese neural network reacted to the USA and the rest of the world

The appearance of Deepseek was a complete surprise for many. The Hill even compared the release of the R1 model with the launch of the Soviet satellite in 1957, after which the space race began between the USSR and the USA. But this time, the United States will have to withstand China, who managed to circumvent export restrictions.

One of the first on Deepseek was the US President Donald Trump.

“The release of the Chinese company by the artificial intelligence of Deepseek should be an alarming signal for American industry that we are obliged to focus on competition. I consider the release of an inexpensive model a very positive event for AI as a whole, because instead of billions of dollars you can spend less and come, I hope, to the same results, ”the American president said .

Trump's closest comrade -in -arms Ilon Musk, who owns Tesla, the social network X and the GROK neural network, refers to Deepseek with suspicion. He considered that

The Chinese company deceived the public and actually used up to 50 thousand video cards and a large amount of money to teach its model.

Also, the billionaire was skeptical of the statements that AI can be created with a relatively small budget.

The founder of Telegram Pavel Durov found that Deepseek's success contributed to a high level of education and competition in China. He cited the example of Chinese students who have repeatedly won the international olympiads in mathematics and programming. According to Durov, the PRC will dominate in the field of technology until the United States has a large -scale educational reform.

Microsoft CEO, Satai Valel, called Deepseek “super impressive” and said that he was very serious about the achievements of the Chinese II industry. The head of Openai Sam Altman, who enthusiastically reacted to the emergence of a new competitor, also responded to Deepseek.

However, on January 29, Openai's attitude towards Deepseek was radically worsened. The American company said that it has evidence about the use of ChatGPT for the training of the Chinese neural network. This is also evidenced by the fact that periodically Deepseek in a conversation with userscalls himself Chatgpt. Trump adviser to AI David Sax openly accused Deepseek of theft of Openai technology.

Perhaps the work of the Chinese neural network will be tried to limit in the United States. So, on January 28, the US naval forces banned the use of Deepseek due to potential risks for the country's security. And the National Security Council is already checking the work of the chatbot. That is how in 2020 the initiative to ban Tiktok in the United States began.

The popularity of Deepseek is not only in Washington. The Italian regulator Garante has announced possible risks, the British government also conducts its investigation. Their fears were not in vain:

On January 29, the WIZ research company discovered the DeepSeek database in the public domain. It contained personal information of users, including the entire history of requests.

It’s too early to say how a scandal with an open database will turn out for Deepseek and its availability outside China. In any case, for China, the dominance of Deepseek in the world media and on mobile platforms already looks like an ideological victory.

What has changed in the AI ​​market with Deepseek

Before the release of Deepseek, the Chat Bott industry developed along the predictable path. Mostly American corporations fought for users who spent billions of dollars on the development of their models. But the situation has changed dramatically. According to The Financial Times, the lead in the AI ​​race will no longer be the most powerful models, as it was before, but the most affordable. Free DeepSeek is already shows the results comparable to paid competitors. This can inspire hundreds of other startups to try to do something similar without huge investments.

The fact that many American and European companies are being prepared to go to chat bots that use open source code of DeepSeek about the change in the industry that many American and European companies are preparing . The desire to use Deepseek was also said in Toyota, Cisco, Yelp and Workday.

Of course, American corporations do not intend to give up. Openai tests the “reasoning” model ChatGPT-O3, which should surpass Deepseek R1. Also, the free version of ChatGPT will work on the O3-Mini model. But when models become available to everyone, it is still unknown.

The Deepseek is also busy with Ilon Musk. It is preparing to release GROK 3, a free neural network for social network X with a mobile application. It is already known now that with GROK 3 it will be possible to talk on complex philosophical topics. But so far it is impossible to say whether the model will be comparable to Deepseek R1.

Perhaps, after some time, a new contender for the championship from China will appear in the neural network race. Alibaba, known for the AliExpress marketplace, introduced its QWEN 2.5 Max model, which now overtakes Deepseek in complex computer tests. Chatbot is available all over the world, including in Russia without VPN. Surely other Chinese corporations will present their AI - for example, Tencent, which owns social networks WeChat and Qzone. She already has a separate unit engaged in neural networks.

In addition, states have yet to decide how to act with respect to Deepseek. So, on January 30, the Italian regulator banned the work of a chatbot in the country due to the lack of transparent information about the use of these citizens. The Irish department has already requested similar information.