Tuesday, the 21st, White House. Donald Trump seems to be rapping sideways as Sam Altman on the stand looks at him with a face that says ze-ostiatan ai-zea. Not every day a project worth $500 billion is presented and Trump has wanted to do so to inaugurate the position of World Sheriff:The new labor association that will promote Artificial Intelligence in the United States is called Stargate. Many have likened it to the Apollo Space Career Project or the Manhattan Nuclear Weapon Generation Program.

“It’s the beginning of a new era,” says OpenAI chief Altman. The idea is to multiply the infrastructure for Artificial Intelligence (AI) in four years, to expand across the country like the massive large-scale data center that is now being built in Texas. He is joined by Oracle founder Larry Ellison and Softbank CEO Masayoshi Son. At the end the photo was taken outside the White House, they are as smiling as they were at a dinner in the squad.

But on the same day, without a cashbox, a Chinese company launches another AA model: Assisted by DeepSeek. The expert leaves his mouth full with the new low cost AA and turns the plans of the Silicon Valley multinationals upside down. But why is that? Moon of Alabama, a geopolitical blog, will publish the following day an entry entitled How Chinese Beat Trump and OpenAI, an article that explains in detail the entire sequence.

“The bubble around Artificial Intelligence, the sterile attempt by the U.S. to monopolize it, and China’s recent counterattack are beautiful lessons that show how innovation should be done, and the U.S. is losing its ability to do so,” they say on the blog. A year and a half ago, when everyone was praising the ChatGPT app, they issued a warning saying that progress was not so much.

A digital bubble

Reading and processing inputs from natural language is not new. In the 1960s, at the Massachusetts Institute of Technology, computer scientist Joseph Weizenbaum interviewed a machine called Eliza through a mainframe or computer. ChatGPT, the initial AI may be more capable than that, its “conversations” may surprise, and in some areas – such as the translation of written languages – it may be practical... But he does not understand what the symbols mean, he cannot solve the mathematical and logical exercises, even if they are simple: “That bubble is not justified,” they concluded in that Moon of Alabama article.

And yet, the bubble continued to swell and swell as the muscles of the gym’s hormonal lackeys grow. OpenAI saw that it could earn billions, and closed its model as a black box to do business. Although the algorithms he uses are known and he takes training content directly from the Internet, despite the enormous waste of money and energy required to process all this data, the U.S. government thought that new technology was born to dominate the world, and to maintain this monopoly he banned the export to China of the most modern semiconductor chips of the company Nvidia.

Now, everything seems to indicate that DeepSeek has sent these intentions and the US business model to take them from the pop. The performance of the Chinese company’s R1 model is as good as that of OpenAI, Anthropic, Google or Meta, but it uses much less computational data for training, since instead of putting into practice all the information it stores, it gives the answer that it considers most appropriate according to the probabilities, as if it were to consult a “group of experts”.

An iPhone of 30 dollars

According to the Reuters news agency, DeepSeek has used 2,000 Nvidia chips for its AA model and spent $5.5 million on it. They are H800 generation chips that are designed to slow down training, older and much slower than the expensive H100 chips used to train the latest version of ChatGPT called o1; but still only 2% of OpenAI chips have been needed by the Chinese company, and their tool is 90% cheaper.

“It’s basically like someone has launched another iPhone-level mobile, but it’s worth $30 instead of $1,000. It is dramatic.” With this comparison, the French businessman Arnaud Bertrand, expert in Chinese affairs, explains the difference between models. Many others have called what happened the “Chinese Sputnik.” On October 4, 1957, the Soviet Union managed to put the first satellite into orbit, and millions of Americans who were terrified by the spectre of the Cold War could see this strange metal ball pass over their heads every 98 minutes; they felt that they had lost their technological advantage over the Soviets until then, just as they have now lost their advantage over China.

In addition, and this is not a small thing, DeepSeek has been expanded in open source by its promoters. In the same LUZ, journalist Gorka Peñagegilano envies China: the trade war is called artificial intelligence as explained in the article “the practice of open source has put the market model in check. Because algorithm chips are expensive, so far only large Silicon Valley companies had the ability to make massive investments and influence the market.” That is, now anyone can use algorithms and this has opened the door to new and cheap AA models.

But DeepSeek doesn’t come alone. The earthquake has caused the Chinese giant Alibaba to have to ramp up the Quewn 2.5 AA application it had in its hands for a long time. “U.S. controls on Nvidia semiconductors to prevent China from advancing in the field of Artificial Intelligence have inadvertently spurred innovation by encouraging companies like DeepSeek, based in Hangzhou, to seek creative solutions,” we read in Moon of Alabama.

And so, on that black Monday in January, those who predicted an investment of 500 billion euros in Washington at sunset, have seen at dawn losses on the stock market of one trillion dollars, as has been the case with Nvidia, which lost 17% in the US Nasdaq index in a single day, 600 billion dollars. The appearance of the new AA model has also solved the skin perfectly to others: Google Alphabet, Microsoft, Meta... all down.

The well-known Marxist economist Michael Roberts, who writes on the blog of The Next Recession, has highlighted in his post AI going DeepSeek what the super-rich investor Ray Dalio has confessed to the Financial Times: those who are trading in these companies have feared that the bubble could explode.

Jevons paradox: more and more

DeepSeek can bring many paradigm shifts, not only in geopolitical relations between the US and China, but also in terms of “cultural warfare.” If we ask the Chinese app about the Tiananmen massacre, it will tell you that it can’t say anything about it; and if you ask ChatGPT about US military secrets, it will tell you the same. We have done another test and asked them to define the journalism of LA LUZ in three words: ChatGPT has chosen the words "truth, transparency, responsible", while DeepSeek has chosen the words "free, critical, independent". Beyond the vague definitions, what is clear is that DeepSeek has come to compete with the Western view that until now was hegemonic in the field of AI.

But we can also clarkly see another thing: because we need less chips, energy, water and money, the growth of Artificial Intelligence will not diminish, but on the contrary. Microsoft chief Satya Nadella mentions the “Jevons paradox” when asked about the earthquake caused by DeepSeek. Economist William Stanley Jevons theorized in 1876 that more efficiency does not mean that you will save material because the consumption rate increases. In LA LUZ we have often mentioned this phenomenon, from the mouth of physicists Antonio Turiel and Adrian Almazan: “This is the case, for example, with the efficiency of cars, when they consume less gasoline: since it is cheaper the cars will be used more and for a longer time, which can eventually lead to a higher level of consumption in the global,” Almazan explained in an interview for the number 2,861.

Something like this can be predicted with the development of AI. According to the International Energy Organization, large-scale data centers in the United States in 2028 will consume 12% of the country’s electricity consumption, which is the same as Japan’s annual consumption. This is how the longing shown by Microsoft, Google and Amazon for the redeployment of old nuclear power plants could be understood.

Will the more efficient and less expensive DeepSeek model change anything? In response to this question, the AFP agency disseminates the words of Andrew Lenders, a professor at Victoria University in New Zealand and expert in Artificial Intelligence, in media such as Le Devoir, France24 in Quebec: “This will help the American giants not to do something 10 times smaller, but to build bigger models, but using the same or more energy as before.”

In search of super-intelligence

So how much does a ChatGPT or DeepSeek response take? To paraphrase the last song by Olaia Inziarte and Natxe de Felipe, we can say that it does not have the same place, it cannot have it, in the hands of a techno-capitalist like Altman. The OpenAI Commissioner believes that having trade-offs in the short term – sacrificing some things to have more things from the other side – is normal... To get to AGI.

The AGI? What the hell is that? The same question was asked by Michael Roberts in his blog post. Artificial General Intelligence is an acronym for the English term. “That’s the Holy Grail of developers,” says Roberts. It means that AA models will become super-intelligent over human intelligence. By achieving this, Altman has pledged that his model will be able to do not the work of an employee but all the tasks of an entire company.” Some see the end of humanity in this career of searching for AGI, but that is the cider of another barrel, fortunately because the social man is more complex than you think.

For Roberts, what’s clear is that Silicon Valley magnets will not stop expanding data centers and producing increasingly advanced chips just because DeepSeek has thwarted their current models. There is a career in Artificial Intelligence between China and the US, who will put the next Sputnik into orbit?