Source: Research by Dong

When Lei Jun wore a brown leather jacket to release new products and became the industry's "price butcher", Huang Renxun, a black leather jacket, once again welcomed Nvidia's stock price plummeted because of the "price butcher" in the industry: a big negative line of 8%, which once again caused market concerns.

NVIDIA's stock price decline this round began with DeepSeek: the same model performance was achieved using only 1/10 of the computing power cost of its peers. The efficiency revolution of a big model has broken the convention that computing power is only large, and has also triggered investors to think:

The computing power required to train a model may be far lower than expected.

A few days ago, Huang Renxun publicly responded to the DeepSeek impact for the first time when he participated in a DNN event. There are three general points of view:

1. DeepSeek is really amazing. (People who don’t believe that domestic big models are really powerful should be able to believe it)

2. Investors’ understanding is completely wrong. "From the perspective of investors, they believe that the world is divided into two stages: pre-training and reasoning, and reasoning is to ask questions to AI and get answers immediately. I don't know who caused this misunderstanding, but obviously this concept is wrong."

3. Post-training is still very important and requires a lot of resources; and reasoning itself is a "computing-intensive part."

These views are not problematic. However, what Lao Huang did not point out is that for investors, the "main contradictions and the main aspects of the contradictions" have changed.

Before DeepSeek-R1 was released, the market's focus was: pre-training and Nvidia. However, after R1 was released, the market's focus changed to: reasoning and low-cost computing power.

In the training stage, the advantages of NVIDIA chips have as high throughput, high parallel speed, etc., all ensure that the company is ahead; however, in the reasoning stage, especially the innovative solutions adopted by R1, the required chip threshold has been reduced.

In other words, it is not that Nvidia products are no longer available, but that other companies' products can also be used.

This contradiction switch will inevitably cause investors to rethink. Even if the company's current and future performance is still excellent.

From this point, technology stock investors in the US and A-shares are the same. Almost none of the technology stocks are driven by valuation. No company has risen sharply because of low valuation, and no company has plummeted because of high valuation.

The core driving is all about: marginal changes and contradiction changes.

So, what will happen to Nvidia look forward to?

In the medium term, with the reasoningExpectations are gradually recognized by the market; if the big model really begins to become popularized on a large scale and penetrates into the industry, it will inevitably stimulate training demand again. Turn market conflicts back to training.

In the long run, every day, Nvidia's risks are getting closer and closer: Huang Renxun retires.

The importance of Huang Renxun’s personal temperament to Nvidia is self-evident. The company's management culture and innovation mechanism are inseparable from Huang Renxun's day and night polishing. If one day, Huang retires and cannot manage the company, his successor at that time is likely to be inferior to Huang Renxun.

The rules of a company's manager are: create a generation with the strongest ability and start the company's glory; then there are important entrepreneurs, and they are also good at entrepreneurship, and they can maintain their careers and develop; then they are either important officials of important officials or professional managers.

Choose a professional manager, which means you are lottery tickets and lucky, which is mixed.

Microsoft missed the mobile Internet during the Ballmer era; under the leadership of Nadella, it ushered in the wave of cloud computing; under the concept of Grove's "only paranoid survive" , Intel created a chip empire, and then under the leadership of professional manager Ou Dening, it missed the wave. The two CEOs of the latter were also bound to move forward and are now becoming the target of acquisition.

So, Nvidia may not be able to get out of this situation.

One other thing is the growth of competitors. Compared with chip companies, Nvidia has two competitive advantages: manufacturing process and CUDA ecosystem.

The former depends on the strength of the chip strategy execution, whether SMIC can withstand the pressure and continue to invest; whether equipment vendors can also achieve the effect of DeepSeek in their existing resources, or directly break through technology.

As long as you spend money, spend time, you have a chance.

The CUDA ecosystem will also take about ten years. If college students learn various courses based on domestic GPUs during college, they will have a little climate.

Back to the current Nvidia, the company also has an important advantage, which is the learning ability and error correction ability of Huang Renxun and Nvidia (derived from the diligence of managers, the company's high talent density, and the system and culture that believe in talents to motivate talents and reward talents). Historically, when Nvidia is in crisis, it often has better cost performance.

Once I asked my friends who worked at Nvidia how to see the company's stock price and whether there are any band operations. His answer is that there are very few band operations, one is that it is inaccurate and the other is meaningless. As long as the company’s colleagues are innovating with extreme diligence, the stock price will definitely rise again.

After all, from a broad perspective, the road to AI entering life has just begun; there are many problems that can be solved by GPU parallel computing.