In recent days, the AI community has been abuzz with the emergence of China's DeepSeek, a groundbreaking large model.

Remarkably, DeepSeek's latest iteration, DeepSeek-R1, demonstrates comparable performance to OpenAI's GPT-4 in tasks encompassing mathematics, coding, and natural language reasoning.

More crucially, DeepSeek-R1 achieves this without relying on pricey NVIDIA chips, boasting a training cost of merely $5.5 million.

This isn't mere hype; there's genuine concern within the US AI community. With DeepSeek-R1 being open-source, AI engineers in the US are avidly dissecting it, attempting to replicate its success in every possible way.

Simultaneously, NVIDIA's position is under threat, as China's achievement demonstrates that advanced AI training can be achieved with just $5.5 million, without the need for NVIDIA's costly H200 or G200 chips—even domestic AI chips suffice.

Many now speculate that the US may have taken a misstep, being lured into a computational power trap by NVIDIA. Billions of dollars were spent on NVIDIA AI cards, creating so-called million-card clusters, blindly stacking chips and power, only to incur astronomical costs without matching China's $5.5 million training achievement.

Consequently, NVIDIA's stock price plummeted by 3.12% in a single day, erasing over $100 billion in market value.

Unlike China's A-share market, where daily fluctuations of 10%, 20%, or even 30% are not uncommon, a 3% drop for a giant like NVIDIA signifies a significant hit, amounting to over $100 billion, or approximately 700 billion Chinese yuan.

This underscores the substantial impact on NVIDIA. If all AI models can be trained cost-effectively, the demand for NVIDIA's AI chips will undoubtedly diminish.

The esteemed journal Nature also noted that US restrictions on Chinese companies accessing advanced AI chips, intended to hinder China's AI development, paradoxically contributed to the success of DeepSeek's R1.

Nature believes that this ban may have inadvertently spurred innovation, providing insights into efficient large model training and even the exploration of human learning patterns, which is both enlightening and shocking.

Industry insiders go further, claiming that in the long run, this could position China as the epicenter of AI research and development, supplanting the traditional leadership of the US.