China's DeepSeek AI is flying past ChatGPT and causes Nvidia stock market loss
Will this change how the LLMs are trained?
2 min. read
Published on
Read our disclosure page to find out how can you help Windows Report sustain the editorial team. Read more
We recently reported about the Stargate Project, developed by a U.S. dream team to create the ultimate infrastructure for artificial intelligence. They might need to step up a bit because a Chinese LLM called DeepSeek is stirring a real craze among users in the U.S. and the rest of the world.
The DeepSeek app recently surpassed OpenAI’s ChatGPT in downloads on App Store after its V3 version release in December 2024. This is a huge hit for OpenAI, but the Chinese AI is not technologically superior even if you look at the DeepSeek’s own benchmarks.
As a matter of fact, OpenAI’s Sam Altman suggested on X that the Chinese copied OpenAI’s model to create their own.
Even so, the users flocked to DeepSeek, probably curious to see the new AI bot king in action. The Chinese were not expecting this popularity surge and now it’s pretty hard to create a new account. Moreover, according to the DeepSeek announcement on the page, they have been the victims of malicious attacks that further slow down the process.
The DeepSeek achievement is, in fact, another. They’ve managed to train their model using almost 1/20 of OpenAI’s budget and with only 2,000 NVIDIA H800 GPUs while OpenAI employed approximately 25,000 NVIDIA A100 GPUs. That is a remarkable achievement, especially after the U.S. chip ban on China.
While DeepSeek-s popularity might be just a temporary bubble, this whole story produced a serious dent in Nvidia’s stocks. According to Blooomberg, Nvidia suffered a $400 million market cap loss, accounting for a 13% drop, the biggest in the U.S. stock market history.
The reason for this huge drop might be that the investors are probably thinking that the DeepSeek achievement will change how the LLM’s will be trained in the future. Even if the Chinese training method is a lot more efficient, it still relies on Nvidia’s chips. The real win here is the fact that LLM development will consume fewer resources, including raw power, which, in turn, will lead to faster advancements.
We’ve learned about this from Neowin. What do you think about DeepSeek and its soaring popularity? Let’s talk about it in the comments below.
User forum
0 messages