Direct Translation via Google Translate. Edited.
As a matter of disclosure, I tried DeepSeek yesterday. It's pretty good, and could give the Big Guys a run for their money.
by Ivan Lizan
[REGNUM] On January 27, the American IT giant Nvidia Corporation lost $600 billion in capitalization. This happened after the Western world discovered the existence of the Chinese startup DeepSeek.
The word "discovered" is not used by chance. DeepSeek has been working since the beginning of December, and by the New Year, specialized publications had paid attention to it. And only at the end of January, business media began to write about the startup.
One can only guess why journalists and stock speculators showed such miraculous slowness. Perhaps they had other things to do in December — they were preparing for Christmas and the New Year. And in January, the US saw a change of power, and the heads of all IT companies immediately swore allegiance to Donald Trump, who had returned to the White House. If the former-current president was an enemy for them during the first five years, now he has become a dear friend — the Democrats have managed to spite the top managers.
It is possible that the "tops" themselves fueled interest in DeepSeek, fanning the flames of the stock market fire. Or someone skillfully played the "bear" (on the downside) on the stock market.
However, this does not really matter, because the US has much bigger problems than the 600 billion virtual dollars that burned. The phenomenon of the Chinese startup has shown the uncompetitiveness of the American approach to the development of artificial intelligence (AI) systems.
THE SANCTIONS WORKED IN THE OPPOSITE DIRECTION
DeepSeek is a chatbot that has proven to be more effective than the super-popular ChatGPT from American OpenAI. In addition, the Chinese version is available for free and runs on less powerful chips.
Chinese specialists managed to create an advanced AI model in conditions where it cannot be created in the United States. The Americans themselves had a hand in this. DeepSeek was created on Nvidia H800 AI accelerators - slower versions of H100 accelerators created specifically for the Chinese market. American sanctions imposed in 2022 under Joe Biden banned the supply of advanced AI accelerators to China in order to slow down the pace of technological development in the PRC.
But, as you can see, instead of slowing down, the US got an acceleration through optimization.
In the end, the need forced the Chinese to use their wits. DeepSeek was trained for $5.5 million, while OpenAI spent $78 million on comparable “pumping” of GPT. That is, the Chinese chatbot was about 90% cheaper. In total, OpenAI spent more than $7 billion on its networks and invested another $1.5 billion in personnel.
The worst thing for the Americans is that DeepSeek distributes many of its developments for free, and asks for the same 90-95% less than OpenAI for paid access to advanced systems - this explains the turmoil in the market. And the DeepSeek model is also open software, which allows other companies to use it to develop their products.
Simply put, the Chinese, with fewer staff and not the most advanced accelerators, managed to create an advanced AI model, and are even ready to provide access to it for pennies compared to what the leader, OpenAI, is asking for.
At the same time, it was an unnamed Chinese company owned by a local billionaire that achieved success, while Microsoft is behind the flagship of American AI. It is its funding that allows it to cover losses. Expenses and losses are fine: they can be written off, especially in the case of Microsoft, the corporation behind OpenAI, known for its impressive graveyard of corporate products. But now OpenAI will have to reconsider its monetization model, reducing prices in conditions when its services do not allow it to break even.
Another major problem for the US is that OpenAI was the country's IT flagship, a kind of dream of a future where a machine can create at the level of a human, creating not only music and images, but also rocket engines - complementing and even replacing humans. AI was supposed to be a tool that would provide the US with global leadership. Only now this leadership can be provided to its competitor - China.
Now it's time to return to Nvidia, whose stocks plummeted on January 27, despite the fact that OpenAI screwed up. The latter company has little to lose - it is not public.
If OpenAI is the software embodiment of the American AI dream, then Nvidia could well be called its hardware base. And it was no secret that Nvidia's stock was wildly overvalued.
At the end of February 2024, Nvidia's capitalization exceeded $1.97 trillion - for comparison, Russia's GDP is $1.92 trillion. And the amount was constantly growing. In September 2024, the corporation's shares fell by 17%, losing almost $279 billion. In January 2025 - another minus 600 billion. But despite this, Nvidia's capitalization is currently $2.89 trillion. They continue to pump money into the huge bubble.
Nvidia shares "fell" because it became obvious that thousands of its most advanced accelerators are not needed for an AI breakthrough, and therefore demand for the company's products may decrease.
GAMES THAT ARE NOT PLAYED
Overall, the DeepSeek story is a good illustration of the general decline in competitiveness in American IT.
Chinese developers have a number of advantages over their US competitors. First of all, it is much cheaper to develop a new product. The average salary of a programmer in California in 2023 was $85,000 per year (7,000 per month), in China - from 18 to 36 thousand (1.5-3 thousand per month). If you multiply the monthly salary by a conditional 100 employees and a couple of years of development, the scale of the problem with the wage fund in the USA will become clear.
Deepseek employed between 100 and 140 people out of the 200-300 employees of the developer company. OpenAI employs more than 3.5 thousand people, which explains the $1.5 billion payroll. It's all about the exorbitant cost of living in California, especially real estate prices. It won't be possible to reduce expenses, and therefore the payroll.
At the same time, the high cost of living is hurting not only the AI industry, but also the American computer games industry, which was hit by the crisis long before DeepSeek got going.
The Chinese game Black Myth: Wukong cost $70 million and was developed over 6 years. It sold over 20 million copies, which brought the developers about $1 billion. Cyberpunk 2077 cost the Polish CD Projekt about 3-4 times more than The Witcher, and sold more than 30 million copies. At the same time, the development cost of Spider-Man 2 from Californian Insomniac Games was $315 million with a circulation of about 11 million copies.
Examples with production budgets and print runs are given for a reason. The goal of any company is to release the best product at the lowest price. The Chinese and Poles cope with this task, the Americans do not. Development is constantly becoming more expensive, the quality of the final product leaves much to be desired, the risks of failure are ever higher.
And this leads to another problem: managers are trying to reduce the risk of losing money. Instead of new projects, they have to choose a well-known franchise, hoping to attract its fans to the ranks of buyers (for example, "Star Wars"). The "agenda" still gets in the way: for example, instead of beautiful female characters, they have to add heroes of "incomprehensible" gender to games, selected by transgender people and other minorities in focus groups.
As a result, instead of an interesting game, we get bland, stuffy and uninspired crafts. Instead of profit, managers record a financial failure, although from the point of view of abstract business they do everything right.
In general, American IT giants have become victims of corporatization and financialization. IT companies in the US are public, their shares are listed on the stock exchange, and IPOs were once used as a tool to attract financing for expansion.
But going public changes both the managers' mindset and motivation. Their number one goal is to increase the company's profitability per share, maintain the stock price, and pay out generous dividends. If this means cutting funding for development, firing a dozen or so engineers, or "optimizing" something (like removing the charger from the smartphone box), you can be sure it will be done.
IT WON'T WORK TO STOP CHINA FROM LEADING
For decades, American IT giants had no competitors as such. This period was their golden age, but the fat years always end. Now the space for competition is expanding, and the number of opponents in the market is growing.
At the same time, we don’t even think about a number of stories of American IT failures. Once upon a time, Android in its version from Google was the only alternative, and depriving a smartphone manufacturer of access to Google services could have been a death sentence for it. Now life without these services has become the norm.
And the American IT sector will only have more problems. It is not clear how to solve them. If we were talking about physical production, it could be moved to Southeast Asia, where many uncompetitive things have been sent. But IT has a different specificity; this industry is the only truly flagship one in the USA. The American authorities, including the federal ones, cannot afford to lose it.
But the problem is that there is nothing to fight the Chinese with. The world games market is global, it will not work to ban the sale of Chinese games. It will not work to ban Chinese companies from developing AI either - they have already tried, the result is known.
All that remains is to selectively ban the use of individual Chinese products under the pretext of ensuring national security. And also to try to quickly "stake out" a part of the once global IT market.
In general, if you can’t win with your head, you’ll have to push with your elbows.
|
