- DeepSeek’s R1 model not only destroys the myth that China can only copy the West, but may even be about to overcome it in the race for AI technology
China’s advancement of artificial intelligence is shaking the foundations of Western dominance in this technological race, evoking comparisons with one of the highest achievements of the USSR.
While investors begin to digest the importance of the IM R1 reasoning model, published fully open source last week, for businesses from large companies such as the chip giant.
Not only does this suggest that the country controlled by the Communist Party has reached the United States, but may also be on the verge of overcoming them.
Continues after advertising
Read more:
The risks of training and inference in more efficient R1 costs also question the thesis that underlies the exorbitant assessments of most actions of the seven magnificent.
Now, Marc Andreessen, a legend of the technological community whose Netscape Navigator has enabled the Internet revolution, is looking for past decades which believes the most appropriate historical parallel is: the day in 1957 where the Soviet Union surpassed the US in space with launch of the first satellite in orbit.
Continues after advertising
“Deepseek R1 is AI Sputnik moment,” posted Andreessen on Sunday (26).
The launch of OpenAi’s chatgpt in November 2022 surprised the world, generating enthusiasm in the AI actions comparable to the “point with” era.
Investors have raised companies from companies such as Nvidia, Microsoft and Tesla in the expectation that a select group will be in a position to raise billions and billions of dollars in AI -related profits.
Continues after advertising
DeepSeek goes to 1st place on the App Store list
Whether it is the Microsoft partner GPT, Alphabet’s Gemini or Claude, projected by Anthropic supported by Amazon, it was unquestionable that a handful of companies had the financial and technological resources necessary to compete in this space.
These companies would monetize their advantage by charging customers by using their owners and closed code models.
Only Elon Musk proved to be able to reach this race by investing billions to build the largest individual computing cluster in the United States in Memphis in record time that will train his new Ia Grok model.
Continues after advertising
However, R1 suggests that this thesis may be wrong.
In a short time, the free Deepseek app has shot the top of the App Store charts.
Your success can impact technology for years.
Your decision to follow the example of the goal, whose model llama is open source, publishing your research, means that anyone can build their own version of R1 and adapt it to their needs.
This can potentially eliminate the technological advantage of AI training that actions from large companies such as Microsoft, Alphabet and Amazon have.
Navidia’s nightmare
For Nvidia CEO Jensen Huang, the news comes exactly when he is increasing the production of his famous Blackwell microchip, a more advanced version of his H100 H100 series, which controls 90% of the AI semiconductor market.
Nvidia circuits are expected to open more than 10% falling, losing the equivalent of $ 350 billion in value when negotiations start.
In an interview with CNBCthe Founder of Perplexity AI, Aravind Srinivas, praised the DeepSek team for reaching the West by employing smart solutions.
This includes halving computing requirements without sacrificing accuracy, changing the most conventional binary coding method – called Floating Point 16 – for the most efficient FP8.
“Usually, the myth is that the Chinese are just good at copying,” Srinivas told CNBC.
“The point is that this is changing,” he continued. “It’s not that China is a copier – they are also innovating.”
c.2025 Fortune Media IP Limited
Distributed by The New York Times Licensing Group
