It was “Black Monday” for American technology companies. Both individual companies and the NASDAQ index ended the session on a considerable minus. Nvidia, the largest AI chip producer in the world, suffered the most. The group’s shares lost almost 17 percent in one day. of its value, which means that the company has sailed out of the company’s capitalization almost $ 600 billion. The “red” session also ended, among others Microsoft (-2.14 percent) and Alphabet (-4.03 percent), and Nasdaq himself shrunk by over 3 percent.
>>> Read also:
Chinese turned out to be “culprit” Startup Deepseek, who presented his model artificial intelligence (LLM) called Deepseek-R1. The new model is to offer possibilities similar to the famous ChatagPT (and in some tests even exceeds it).

However, real anxiety among investors on Wall Street was caused by the fact that the creation and training of the Deepseek-R1 model was to cost only a fraction of what American companies from the so-called They allocate the magnificent seven each year on investments related to AI.
How much did the Deepseek-R1 cost? Certainly not $ 6 million
Deepseek creators say that to train your AI They spent less than $ 6 million. This amount seems ridiculously small.
The problem is that Deepseek declarations probably don’t have much to do with the facts. Let’s start with the fact that the Chinese startup did not come from nowhere (although he is trying to create such an impression). Its creator, Liang Wenfeng has been associated with the Tech industry since at least 2015, when he founded the Hedge Fund High-Flyer. 40-year-old Venfeng, who is sometimes called “Chinese Altman” in the media, is today one of the greatest AI specialists in the Middle Kingdom.
It is also not true that the Deepseek-R1 model has been trained only 1000 NVIDIA A100 cards. , in 2019, Wenfeng had such a number of units, but in 2022 (even before the entry of American sanctions, which limited the possibility of buying these cards by Chinese companies) their number exceeded 10 0000. As experts estimate the experts quoted by the BBC portal, Currently, there are 10 to even 50,000 NVIDIA units in High-Flyer hands.

So why did 1000 NVIDIA cards and 6 million dollars come from in media messages? These numbers – concern only the less advanced Deepseek -V3 model, for which the Chinese startup used cheaper NVIDIA A800 units.
Even if we assume that the Deepseek-R1 model has been trained On “only” 10,000 A100 cards, the estimated cost of building such a cluster is from around 100 to 150 million dollars. The story of a young Chinese start, who, having a few million dollars, challenged American giants in their hands, can therefore be treated in terms of urban legend.
Did Deepseek really build his openai for $ 6 million? Of course not. It also seems absurd to believe that innovations implemented by Deepseek are completely unknown to the leading AI researchers in other laboratories in the world
– emphasizes Stacy Rasgon, an analyst at the Bernstein fund, cited by.
Experts note, however, that even if Deepseek has allocated not a few, but hundreds of millions of dollars for the construction of their models, this amount is still a drop in a sea of global expenses on AI. It is estimated that American companies from the so -called A great seven for the development of artificial intelligence spend about 250 billion in a yeare. $ 500 billion is the budget of the project announced by President Donald Trump Stargatein which, among others Openai, Oracle and Japanese Softbank.
The effectiveness of the Deepseek-R1 model is also important. As the Chinese company emphasizes, depending on the task performed, Using it is 20 to even 50 times cheaper than in the case of a competitive OPENA1 O1 model.
In this context, the fear of some investors, whether technology companies do not burn money for AI investments by accident, seems to be understandable. Can the Monday session be the first trailer of a cracking bubble? Before such a scenario, he warns, among others Ray Daliocreator of the Bridgewater Associates fund and one of the legends of Wall Street. According to Dalio, the current situation is more and more like 1998-1999, That is the peak of the bubble “dotcom”, when the valuations of many online startups were completely detached from their foundations.
A new, important technology has been created, which will definitely change the world and succeed. The problem is that some are confusing with the success of the investments themselves
-.
Americans can also be disturbing the declarations of Liang Wenfeng himself, who in July 2024 in an interview with the Chinese periodical China Academy outlined China’s power plans related to AI.
Chinese companies have been using technological innovations developed elsewhere for years and earned on them through applications, but this is not enough. This time our goal is not profits, but to move the technological limits to drive the entire ecosystem
–
Altman himself praises Deepseek-R1. “Impressive”
Interest in the application on Monday Deepseek (works on the V3 model, not R1!) It was so big that At some point, the company had to introduce registration limits for new users. The Chinese startup also announced that he fell victim to hacker attacks on his servers.
Interestingly, Altman himself, the co -creator of Opeli, does not seem to succumb to the panic digging investors in Wall Street. He speaks positively about Deepseek, but he ensures that his company is already working on much better models.
The R1 model from Deepseek is impressive, especially considering what it is able to offer for the price
– Altman himself wrote in the entry posted on the website X. “We are glad that we can continue to implement our research plan and believe that now – more than ever – we need more computing power to accomplish our mission” – added the general director of Opeli.
Source: Gazeta

Mabel is a talented author and journalist with a passion for all things technology. As an experienced writer for the 247 News Agency, she has established a reputation for her in-depth reporting and expert analysis on the latest developments in the tech industry.