This week it has not started too well for the technological sector related to AI in the parks. For sample, a button: Nvidia The session in the US bags fell with force as soon as possible, with their actions reaching descend halfway around 17% and losing about half a billion dollars.
It is not the only technology related to the chips that you are experiencing more than notable falls at this start of the week. At half a session in the US bags, Micron fell above 12%, Marvell around 17%, TSMC registered a decrease close to 15%, Broadcom lost more than 18%and AMD almost 6.5%. Even European as ASML have lowered hard on the stock market. What has caused this cataclysm? Well, nothing less than the launch of a family of AI models in Chinaand the results that it has obtained with respect to others developed by Technologicals of the West. Where did it come from? From a Chinese startup called DeepSeek.
It all started on January 20 with the launch of a reasoning model called Depseek-R1, with an associated chatbot. Shortly before, at the end of 2024 another, Deepseek V3, a large model of language. Then he already raised some suspicions, since his creators decided to be Open Source, in addition to ensuring that he developed in less than two months and that somewhat less than 6 million dollars were invested in their development. It is, therefore, a much smaller amount than what western companies and investors invest in developing and training this type of models.
Since then, the runrún in the technological sector related to AI on what was achieved with this model has been increasing, reaching its highest point in the late last week when new data on the performance of Deepseek-R1 and other models appear and other models and other models of the company, as V3. Also on the necessary hardware for its creation, in principle chipsets and components much less powerful than those used by other US companies to create their own. Many are not yet explained by Deepseek and how they have done so, and the reaction of the bags has not been long in coming.
If it is true, what is raised by Deepseek, it would not be necessary Team to work with AI. For now, software and IA development companies that quote on the stock market, although many record losses, maintain the type, although we will have to see what happens in the next few days.
Where exactly Deepseek comes from?
As we mentioned, Depseek is a China startup, which has more than a year, and that ensures that with their AI models that can offer similar performance, or even higher In some aspects, to the chatbot and more powerful models developed by American and European entities. In addition, they have apparently for a much smaller cost than those developed by other companies.
As if this were not enough, the creators of their models, according to Bloomberg, ensure that the training of the models have been less complex, and that in their development, training and model are consumed much less energy than with models of other companies.
Founded by the person in charge of the Investment Fund focused on the High-Flyer, Liang Wenfeng, makes its Open Source models in order for the developer community to examine their code and improve them. The apps that have launched, with Deepseek V3 integrated, have raised great interest, and the one that has been developing for iOS is already in the first position of the iPhone downloads in the United States, China, Singapore, Canada, Australia and the United Kingdom in just A few days.
In the Android store he also occupies the top positions in several countries. Among them, in Spain, where it is the second application in downloads at this time. On January 25, more than 1.6 million times had already been downloaded worldwide. Su chatbot differs from others in what it is capable of articulating your reasoning before offering an answer to a promptmatching and even surpassing Chatgpt in some aspects.
Advantages that Deepseek’s models would have with respect to others
As we have mentioned, the Cost of Development and Training of Depseek models would be just a part of what is needed to do the same with the best and most powerful models and chatbots of OpenAi and goal. Therefore, Its energy efficiency level is much greaterand questions the need to be necessary to invest large amounts of money to move forward in the development of AI models.
Also, since in theory I would not have the latest chips of the developer of components and accelerators of the last generation by the embargo imposed by the United States on the sale of chipsets and machinery to manufacture them to China, it would demonstrate that the last models of chips and accelerators to advance in Ia. Nor are the most powerful. Something totally different from what companies dedicated to the manufacture of hardware for AI, but above all the companies in charge of the development of generative AI systems.
According to Sources from Deepseek, R1 is very close, and even exceeds, the results of other AI models in various specialized test banks. Among them are Aime 2024 for mathematical tasks, MMLU for General Culture and Alpacaeval 2.0 for performance with batteries of questions and answers.
In addition, it is in the top positions in a performance index affiliated with the University of Berkeley, known as Chatbot Arena. As to V3, managed to overcome models such as GPT-4O or call 3.1 405b. Of course, his chatbot seems to have some important gaps when asked about controversial issues in China, as on the other hand could wait.
Are there reasons for the alarm in the US and among chips manufacturers?
Not having allegedly needed last generation chips to advance in development of AI, Deepseek has put in Solfa the restrictions imposed by the United States on the sale of semiconductors and machines for the development of last generation chips.
The advances of Deepseek, which already has several models available, would also demonstrate the ability of China’s engineers, who are being managed to overcome, or skip, restrictions. They would have achieved it with a very simple premise: to improve the efficiency of the models with the limited resources available.
Your Open Source model is also designed to get the greatest number of users as soon as possible, possibly and then develop monetization strategies to take advantage of the user base you get. And by offering more affordable models, it has already managed to reduce the costs of AI developers in China, which has made many join that base.
Deepseek’s success can lead to OpenAI and other AI suppliers, especially Americans, have to significantly lower their prices to maintain their leadership. In addition, it opens a debate around the enormous amounts of money that both companies and government institutions are investing in AI so as not to be left behind.
Less than a week ago, in fact, Trump announced a huge investment in the start -up of the Stargate project, which aims to turn the US in the world leader into artificial intelligence, and in which they plan to invest up to half a billion dollars . Meta also has just announced another powerful investment in AI, especially for infrastructure: 65,000 million. Much more than what Deepseek points out that it is needed to develop and train AI systems.
In view of what this startup promises, the reaction of the bags as soon as information and possible results of the company’s models have been made, it has been fled to disrupt the values related to the development of chips for ia , as well as some of the companies that are directly related to their development, although to a lesser extent.
We will have to see how markets evolve in the next few days to see if today’s falls become a crisis in the development of chips for AI, or if everything stays in a scare for technological ones.
But if it is confirmed what Depseek ensures about the creation and set -up They will have much more complicated to make their message cale in the world of technology. Above all, If there are other Chinese companies that are working on AI models and systems with a similar approachand that they are about to see the light.
