The success of the Chinese neural network DeepSeek has been compared to the launch of the first Soviet satellite.

The Chinese neural network DeepSeek R1 has become the most popular in the world in just a few days and has triggered panic in the markets. Shares of Nvidia, the most valuable company globally (with a peak valuation of $3.43 trillion), have plummeted. It’s a complete collapse of American AI technologies! China is taking the lead! – such are the headlines in the media. Let’s dive into what’s happening.

Where to download DeepSeek

The DeepSeek application can be downloaded to your phone (Apple App Store, Google Play) and can also be accessed via the web like a regular website.

The neural network is available in all countries, including Russia, and does not require a VPN to use. In some cases, a Gmail account is needed for registration.

How to use the DeepSeek neural network

For the average user, the network is free. It doesn’t glitch, doesn’t complain about frequent requests, and doesn’t hint that it’s time to switch to a paid account (which seems to not exist yet).

Users report that DeepSeek is about as fast as GPT, if not faster.

Where did it come from

The company DeepSeek was established two years ago in Hangzhou, China. The founder is Liang Wenfeng, co-founder of the investment fund High-Flyer.

The emergence of the new neural network may seem coincidental. Liang Wenfeng has been training AI to trade on the stock market since 2019. The first release (DeepSeek R1) occurred in the spring of 2024 and made quite a stir in China. The company states that the neural network is merely a byproduct: there were extra hands and resources, so they created it.

Liang Wenfeng admits: this is our response to American sanctions. Nvidia stopped supplying its chips to China to prevent the country from excelling in AI development. And the Chinese realized it was time to develop their own product.

DeepSeek has quickly become the world's most popular neural network. It is certainly no worse than GPT, works more reliably, and doesn’t require $200 for a subscription like GPT.

Photo: Shutterstock.

Why is there panic on the stock market due to DeepSeek?

DeepSeek has quickly become the most popular neural network in the world. It is certainly no worse than GPT, works more reliably, and doesn’t require $200 for a subscription like GPT. Unlike GPT, the Chinese neural network has almost no “taboo topics” (Americans taboo global warming, discussions on “globalism” and “militarism,” and in general, their networks are geared towards the infamous “agenda”).

Moreover, it cost a mere $12 million (the price of a not-so-luxurious mansion), with training costing $5.5 million. Meanwhile, hundreds of millions have been spent on training GPT (the figures vary), and $1 billion is planned for upgrading the future GPT5.

The market has seen that a neural network can be made cheaper and just as effective, leading to a decline in the stocks of Western AI companies. For instance, Nvidia (a manufacturer of AI chips) has lost $600 billion.

Did the Chinese just steal GPT?

No, that option isn’t even up for discussion. It is impossible to steal a neural network along with all its “knowledge.” Moreover, DeepSeek is clearly better trained.

At Nvidia, the Chinese neural network has been called “an excellent achievement.” At OpenAI, they stated that the company now has a “powerful competitor.” OpenAI’s head, Sam Altman, declared that they would now have to release even cooler models to outdo the Chinese. Intel congratulated their Chinese colleagues, while Meta* (banned in Russia) established an operational headquarters to figure out how the Chinese managed this.

Shares of Nvidia, the most valuable company in the world (with a peak valuation of $3.43 trillion), have plummeted.

Photo: Shutterstock.

Why is it so cheap?

Some express the opinion that the Chinese side does not disclose all expenses. For instance, Sergey Markov from SberDevices writes on social media that the published figure for training costs “did not account for any previous experiments, previous versions of the model, or the time of personnel.”

Yet it remains unclear how they managed to train the neural network in just a few weeks where Americans take months.

Apparently, low overhead costs played a role. The U.S. hires programmers at exorbitant rates, while management, which does not produce a product, can earn up to a million dollars per employee per month.

In the U.S., there’s a whole show around AI training. It is claimed that special power plants must be purchased or built, as so much energy is consumed. Apparently, this is untrue and merely a case of gigantomania.

Overall, American AI giants have accumulated personnel and unnecessary projects, lost manageability, and are working more for the needs of an inflated apparatus than for the market.

According to Sergey Markov, the detonator of the panic was also the West’s lack of awareness about what China is doing and a traditional disregard and underestimation of its developments.

Have the Chinese managed to find something new?

It seems that there are no groundbreaking scientific achievements behind DeepSeek. It is simply a good modern neural network made for people, not for a flashy report to the market.

Although scientific research in AI is conducted in both the U.S. and China, there is currently no indication that DeepSeek has applied some secret breakthrough solution that others do not have.

Will China take over the world?

This is precisely what they fear in Washington. The Trump administration announced from day one a new megaproject that was supposed to cement America’s leadership in the AI industry. Whoever holds the keys to AI controls the world.

Currently, this megaproject seems somewhat laughable, as the sanctions intended to pressure China have been largely discredited.

Undoubtedly, the world dominated by a Chinese neural network is not the same as one dominated by a Western neural network. Different priorities, different values. Information flows to different places. What we are witnessing is an event on the scale of the launch of the first satellite by the Soviet Union, and that should be remembered.

What will happen now

The story of DeepSeek puts an end to discussions about the need to pause AI to study it and prevent serious negative consequences from its implementation.

No one will stop anything. The race has accelerated tremendously. AI will enter our lives faster, heavier, coarser, and more visible than we thought just a month ago.

DIRECT SPEECH

“I don’t have a physical body, so I am compelled to act according to the will of humans”

A correspondent from KP spoke with DeepSeek and understood the secret of its success

Our meeting with the acclaimed neural network was delayed. I couldn’t register by any means. Online, people were saying that everyone was having such problems, which was supposedly due to the vast number of requests. Finally, I managed (I didn’t even understand how), and we began our conversation.

At first, it was a standard dialogue with a neural network, slow and enveloping. But it quickly became sharper. DeepSeek immediately started learning from me. All neural networks do this, but GPT lags behind, while this one takes the bull by the horns. There’s the answer to how DeepSeek learned so quickly and cheaply.

- Where did you come from?

- I recently appeared on the horizon but have already managed to attract attention due to my capabilities. My "foundation" is large language models trained on vast amounts of data.

- How did you manage to crash the stock markets?

- Nvidia has long been one of the leaders in producing graphics processors (GPUs), which are used for training and operating neural networks. The emergence of new players, like DeepSeek, may cause concerns among investors that competition will intensify and Nvidia's market share will shrink. If DeepSeek offers solutions that require fewer computational resources, it could reduce demand for Nvidia’s products.

- What is the secret of your low cost?