DeepSeek Rises: New Opportunities in the AI Agent Track

The emergence of DeepSeek can definitely be considered a "black swan" event in 2025.

This black swan directly "blackened" the U.S. stock market and broke the conventional thinking regarding the subsequent development of artificial intelligence.

Previously, the former CEO of Google stated that the competition in artificial intelligence is a race with a hundred billion dollar threshold; without crossing this financial barrier, one cannot even obtain a ticket to the competition.

Zhu Xiaohu from Jinsha River Venture Capital once mentioned that he completely avoids investing in domestic large model companies because he believes that domestic companies can never catch up with the U.S. in large models.

The reason these two individuals have such assertions is that there has always been a fundamental assumption in AI competition: to win in the competition, especially in the competition for foundational models, computing power and funding are insurmountable barriers.

Our country's enterprises are stuck on computing power—unable to obtain the best NVIDIA graphics cards.

The most direct shock brought to the world by the emergence of DeepSeek is that it achieved outstanding performance at an extremely low cost.

This achievement directly dropped the biggest "shock bomb" in the current global competition in artificial intelligence:

If performance improvements and cost reductions can still be achieved through algorithmic modifications and optimizations, then is such a massive investment in computing power and energy still necessary to continue advancing and developing AI applications?

Thus, the founder of "Black Wukong" praised DeepSeek as a national-level achievement.

However, no matter how excellent DeepSeek is, it is still a result developed on the shoulders of giants, as Yang Lekun said:

This is the victory of open-source large models over closed-source large models.

I believe DeepSeek has three very profound impacts on the subsequent development of artificial intelligence:

First: It greatly reduces the cost of large models, which means countless small and medium-sized companies may join the competition in AI large models and develop specialized large models in various niche fields and industries.

Second: The wave of AI entrepreneurship in China will explode—far beyond what we see now.

Third: Chinese entrepreneurial teams in the AI track will be viewed with new eyes, likely reigniting global venture capital interest.

The success of DeepSeek means that the computing power barrier is not as high as imagined. Since DeepSeek went online, many laboratories, including universities in Hong Kong, have successfully replicated DeepSeek's results.

We can easily imagine that a large number of companies will further optimize models based on DeepSeek's open-source foundation, leading us to train good large models even with less top-tier chips.

The first impact is already interfering with the U.S. stock market.

Due to the astonishing results demonstrated by DeepSeek in terms of cost and performance, core companies in the AI track are now facing reevaluation.

NVIDIA dropped over 10% at the opening on Monday. Other related companies are facing similar situations.

I immediately thought of Tesla.

Previously, I believed that Tesla's visual solution was likely the ultimate solution for autonomous driving. In this solution, Tesla's advantages lie in large models and data.

Now with the emergence of DeepSeek, Tesla's advantage in large models may not be as obvious, leaving only the data advantage still far ahead.

But how long this leading advantage can be maintained will depend on the subsequent development of other electric vehicle companies.

Returning to the crypto ecosystem.

On the surface, the emergence of DeepSeek has pulled down the U.S. stock market and also affected the crypto ecosystem, especially the AI agent track. Many people believe that the AI agent track has collapsed. However, I think this actually indicates from another perspective that the competition in the AI agent track is far from over.

Because if large models can iterate at such a fast pace, it means that the leading advantages achieved by existing AI agents are only temporary; if they cannot keep up with trends and technological iterations in the future, they could be overturned by newcomers using the latest technologies and models at any time.

At the same time, this also means that the research and development threshold for AI agents has further decreased, allowing more participants to engage in the development of AI agents, which also means that local Chinese teams have a complete opportunity to shine in this arena.

Today is the last article of our Year of the Dragon. After the Spring Festival of the Year of the Snake, we will meet again. I hope that in the new year we can continue to maintain our enthusiasm and continue to find our love and gain in this industry.