#打榜优质内容

China's artificial intelligence DeepSeek disrupts the markets: 1 trillion dollars could evaporate

Nvidia, leading the explosion in artificial intelligence stocks with its chips being seen as the backbone of artificial intelligence technology, saw its shares drop 10 percent in pre-market trading on Monday, the California-based chip firm.

Chinese artificial intelligence laboratory DeepSeek's large language model, named after itself, (LLM), has become one of the biggest competitors of the American firm OpenAI's ChatGPT, shaking global technology stocks on Monday.

The performance of the recently released DeepSeek-R1 language model, which is one of the latest models of ChatGPT with fewer chips and costs, at the o1 level caused Nvidia, famous for its artificial intelligence chips, to lose value. (LLM)

It is believed that ChatGPT requires 10,000 Nvidia graphics processing units to process its training data. DeepSeek engineers say they achieved similar results with only 2,000 GPUs.

Nvidia, with its chips being seen as the backbone of artificial intelligence technology, had led an explosion in artificial intelligence stocks. Shares of the California-based chip company fell 10 percent in pre-market trading on Monday.

By the way, DeepSeek's mobile application surpassed ChatGPT on Monday and became the highest-rated free application on the Apple App Store in the United States.

On the other hand, according to Reuters, the application in question works with the previous model of the initiative, DeepSeek-V3.

Data research firm Sensor Tower states that the model released on January 10 has gained popularity among U.S. users.

The recently released model DeepSeek R1 is particularly praised by researchers for its ability to perform complex reasoning tasks in mathematics and coding. The model uses a similar 'train of thought' approach to ChatGPT o1, allowing it to process queries step by step and solve problems.

An open source model
Moreover, DeepSeek-R1 has been largely released as open source. So anyone can access the codes of the vehicle and customize the LLM using these codes. The training data, however, is patented.

On the other hand, OpenAI had released the o1 model privately and is currently selling it only to users with monthly packages ranging from $20 to $200.

Zihan Wang, a DeepSeek employee speaking to MIT Technology Review and a computer science doctoral student at Northwestern University, says China has had to resort to innovative ways to counter sanctions.

According to this, DeepSeek had to rearrange the training process to create the R1 model. Because the performance of the chips that the chip firm Nvidia is allowed to sell for the Chinese market is half the speed of the company's best products. Therefore, China has rearranged the training process to reduce the load on the (GPU) graphic processing units it purchased from Nvidia.