Text & Image | Tang Jie

Reviewing the trends in the US stock market over the past week, several "suspicious points" warrant further discussion.

Undeniably, DeepSeek has thoroughly shaken the global tech community. Since the launch of its "low-cost, high-performance" DeepSeek-R1 model on January 20, it swiftly ascended to the top of the IOS download rankings in over 140 countries, becoming the most popular Chinese AI application globally in record time.

Moreover, on January 28, it triggered the largest single-stock correction in US stock market history, with NVIDIA's share price plummeting by $550 billion in a week. The entire new US administration, including Trump, was caught off guard and began claiming that DeepSeek illegally "distilled" ChatGPT and demanded that the Singaporean government investigate whether DeepSeek had acquired AI chips "banned" by the United States.

However, what truly merits attention is that following the turbulence caused by DeepSeek, the US stock market swiftly absorbed the declines and quietly shifted sectors. The Nasdaq Index gradually rebounded from around 19,200 points on January 23 to hit 19,969 points intraday on January 31.

Without delving into the disruptive changes DeepSeek has brought at the technical level, we believe at least two key trends for the year are emerging:

Firstly, AI is undeniably the representative of future global "advanced productivity" and the most critical direction of technological development. Previously, some market voices questioned AI, believing its progress over the past few years has been slower than previous revolutionary technologies (like the internet and semiconductors) and that it couldn't meet high market expectations. However, DeepSeek's significant advancements show that AI deserves its reputation as a revolutionary technology, and its progress will accelerate.

Secondly, significant technological advancements occurring outside the United States have limited influence. On one hand, domestic tech giants are the primary participants in the US, and it's unrealistic to expect the US government and core enterprises to adopt China's large models. On the other hand, these giants already have excessively high market values, so the market now has even more reason to focus on AI infrastructure, AI security, and AI application areas likely to obtain "more orders".

01 Join if You Can't Compete

DeepSeek didn't emerge out of nowhere, and R3 isn't QuantOpt's first extremely "cost-effective" large model product. Since ChatGPT ignited the entire AI concept, DeepSeek has been a player in China's large model field and had already launched the DeepSeek Coder V2 model in May 2024, achieving GPT-4's effects at an extremely low cost.

If we must compare, the technological progress at that time was not smaller in magnitude than it is now, but its market impact is incomparable. After that launch, DeepSeek continued innovating along the path of "low cost, high effectiveness" until the release of DeepSeek-R1 in late 2024, officially becoming a Chinese large model product with global influence.

For six consecutive months, new products were released monthly, with each making significant progress, achieving lower costs and better effects. This is what revolutionary technology should look like in its early stages of development and is the fundamental reason DeepSeek has caused tremendous shock in the US stock market:

These powerful and wealthy large companies in the US and the large models they support have spent hundreds of billions of dollars on purchasing chips and building computing centers but have never achieved progress at the speed of DeepSeek since 2024.

It's worth noting that Silicon Valley isn't short of innovators or inflated egos. Leaders of American tech companies, like Musk, habitually portray themselves as visionaries, predicting future trends decades in advance, and past facts have proven this. Therefore, investors grant these companies generous returns, regardless of whether their ideas succeed or not. But "spoiling" isn't entirely beneficial; it makes tech elites believe they're world leaders.

The long-term defamatory reports of mainstream American media on China have also narrowed their understanding of China's technological innovation. In their eyes, Chinese tech companies lack the same innovative capabilities as American companies, and China doesn't have the soil to encourage technological innovation. The opposite is true; after years of internal strong regulation and external sanctions, the energy for innovation in China's tech industry hasn't diminished but is still thriving.

On one hand, there's blind self-confidence; on the other, there's ignorance of competitors. The combination of these two factors creates a natural and huge information gap. In the capital market, the information gap is the biggest expectation gap, and once this expectation gap is discovered, it will be quickly "filled".

Meta's Recent Performance: Recently, Meta, also belonging to the open-source model camp, released its latest financial report, with revenue increasing by 21% year-on-year, net profit of 43%, and an overall operating profit margin of 48%, which is quite impressive. However, the market's true focus is on two things: first, the intensity of capital expenditures for the entire year of 2025; second, whether DeepSeek will surpass the company's Llama and become the global leader in open-source models.

Regarding the first point, Zuckerberg's recent stance has been very firm; he believes it's too early to slow down investment in computing facilities, and investment in AI-related projects in 2025 will reach $65 billion, expected to be more than twice that of 2023. Regarding the second point, Zuckerberg praised DeepSeek's achievements and also expressed Meta's determination to continue investing.

Additionally, according to some news descriptions, Meta has already established an internal team dedicated to analyzing DeepSeek, placing this Chinese startup in the same competitive position as OpenAI and Google. Essentially, Llama, also an open-source large model, can more easily draw on DeepSeek's successful experience, reducing Meta's own R&D costs.

Therefore, under DeepSeek's impact, Meta's share price was much less affected among the seven giants of the US stock market. "Join if you can't compete" is the success code inscribed in Meta's genes; and precisely because of this, in an era of "flourishing" large models, the launch of computing center projects will not only not slow down but will accelerate against the backdrop of geopolitical competition.

02 Changing the World

Thinking deeper, what truly matters is whether AI has promoted productivity improvement and how these value enhancements are distributed. Judging from current results, the effect on productivity improvement is relatively weak, and there's still a huge gap between ideal and reality.

Of course, the disconnect between new technologies and the real economy isn't new; from personal computers to electric vehicles, technological revolutions often take decades to be reflected in productivity data. But no matter what kind of revolutionary technology it is, it must demonstrate that the gap is constantly narrowing; otherwise, the bubble hyped by investors will quickly burst.

Therefore, another significant impact of DeepSeek on the entire market is that it has extended the life of the AI bubble. As everyone knows, one of the major concerns about AI in the market used to be its high cost, making it affordable only for wealthy and powerful large companies; but now, DeepSeek is rapidly driving down costs, meaning the path for AI to impact actual productivity will be significantly shortened.

Apart from DeepSeek, in 2024, more and more C-end users came into contact with and used AI. According to statistics from the Federal Reserve Bank of St. Louis, more than 40% of the US population used generative AI in 2024; compared to personal computers and the internet, the pace of generative AI's popularization in the mass user market is also far ahead.

On the Chinese side, although there are no corresponding statistics, with the widespread adoption of large-scale mobile internet and social media applications, generative AI has affected our lives at various levels.

In summary, with the accelerated cost reduction at the technical level and the accelerated increase in popularization at the user level, more and more industries will proactively accelerate their contact and integration with AI in 2025, either providing services for AI and becoming a source of its rapid growth or being empowered by AI until becoming an AI-native enterprise and AI-izing all service capabilities.

Along these two directions, we find that compared to the "seven giants" of the US stock market, which have already soared, there's still much untapped potential in the market value of many small and medium-sized enterprises upstream and downstream of the AI industry chain, with plenty of room for growth.

For example, Credo (CRDO), a semiconductor company, connects computing/data infrastructure to the internet or cloud by designing Ethernet chipsets, focusing on higher bandwidth, power consumption, cost, security, reliability, and end-to-end signal integrity requirements. Simply put, Credo is part of the data/cloud supply chain; they don't design computing chips like NVIDIA but focus on data transmission chips.

The strong data transmission demand of AI computing centers has drawn market attention to products from such companies. The company's financial report showed a year-on-year increase in revenue of 64% and a quarter-on-quarter increase of 21% for the second quarter of fiscal year 2025 (Q4 2024); earnings per share reached $0.07, also significantly exceeding expectations, making it the most surprising quarter in the past two years. It's worth mentioning that the company's share price was significantly affected by DeepSeek, falling to $69, a significant discount compared to the previous high of $86, making it worth tracking continuously.

Among the companies supporting AI, there are also some outstanding representative enterprises. For example, CLS, which provides computing center server integration solutions; Argan (AGX), a private enterprise engaged in power generation engineering EPC services in the US, and so on. These companies share a characteristic: they have good business prospects against the backdrop of large-scale launches of AI computing centers.

Another type of company benefiting from AI is at the application level. Among such companies, Palantir is still the flagship bearer. As a representative of AI enterprises in the military field in US stocks, Palantir has a high degree of scarcity in the entire US stock market; at the same time, they also have the most prominent AI application scenario with the strongest payment ability—the military. With large models being able to rapidly reduce costs, the company's business and performance will have significant room for improvement.

Similarly, there's Q2 (QTWO), an AI enterprise that has already partnered with 50% of the top 100 banks in the US; of course, there's also Meta, which benefits from AI advertising bonuses, as mentioned earlier. In fact, the acceleration of AI applications has already formed a trend last year, and the gains of many enterprises are already on the high side. In this case, waiting for a correction will be a good strategy.

03 Conclusion

If we think deeper, is the advent of the AI era the best of times or the worst of times?

When AI truly enters various industries on a large scale and affects the operation of actual productivity, will it lead to more unemployment? With hundreds of billions of dollars invested in development costs, how much money will ultimately be able to enhance value and efficiency? After DeepSeek brings about large-scale cost reductions, will there be even bigger "scares" ahead?

No one truly knows the answers to these questions, just as no one knows what will happen tomorrow. But one thing is clear: the technological revolution is accelerating the release of its potential, and each of us must be prepared for it, both in terms of skills and cognition.

Disclaimer: This article is for learning and exchange purposes only and does not constitute investment advice.