Will DeepSeek rewrite Nvidia's growth script?

Source: FT Chinese

In an online event last week, Huang Renxun had a conversation with the CEO of DataDirect Networks. During the conversation, Huang Renxun responded for the first time to the impact of DeepSeek on Nvidia.

Almost a month ago, DeepSeek released the open source version of its R1 model, after which Nvidia's stock price was like a roller coaster, with the stock price plummeting 17% at one point, causing Nvidia's market value to evaporate by nearly $600 billion in three days.

Almost a month later, Nvidia's stock price has basically "recovered lost ground." Huang Renxun also mentioned DeepSeek for the first time. He said the release of R1 was "incredibly exciting" and said that the previous market reaction was a misunderstanding. "I think the market's reaction to R1 was 'Oh my god, AI is over! It's coming down from the sky, and we don't need to do any computing anymore!'" Huang said, "and the opposite is true." Huang believes that the launch of the DeepSeek R1 model has accelerated the popularity of AI, which does not mean that the market no longer needs computing resources, but on the contrary, driven by DeepSeek, the market will need more computing power in the future. Huang chose a clever time to speak out and talk about the impact of DeepSeek. First, Nvidia has almost recovered from its stock price plunge; second, Nvidia will release its fourth-quarter earnings in a few days. For the market, Nvidia's earnings have become as important as key economic data. Market analysts will judge the prosperity of the AI ​​market based on the data, and Nvidia's stock price will fluctuate accordingly. But this time, market analysts seem to have a lot of disagreements about Nvidia's earnings and stock price outlook. The emergence of DeepSeek means the possibility of low-cost large model training methods. This breaks the moat that the giants have built with a lot of money, chips and computing power. In the training of large models, if the expected performance can be achieved with relatively less computing power, then the future market does not need so many high-end AI chips at all. Does Nvidia's stock price need to squeeze the bubble?

From this point of view, the rise of DeepSeek has led to the plunge of Nvidia, which is a relatively direct market feedback based on this analysis. Whether this is a "misunderstanding" may take time to verify.

But one month after DeepSeek suddenly became popular before the Spring Festival, everyone's judgment on the market is changing. As Huang Renxun said, the release of DeepSeek's new open source model has set off a new wave of AI, and the scope of AI application is expanding, which means that more computing resources will be needed in the future.

Analysts who hold this view will use the Jevons paradox in economic theory to summarize this counterintuitive phenomenon of resource efficiency: when resource utilization efficiency is improved, it may actually lead to an increase in resource consumption rather than a decrease. Computing power is also a resource. Driven by technological progress, computing power efficiency is improving and the market is further expanding, which in turn has led to a continuous increase in market demand for computing power.

This is being gradually verified by the market. According to a report by the Economic Observer a week ago, more and more Chinese and foreign companies have accessed DeepSeek, and the rapidly growing number of customers has also put great pressure on the servers of various platforms. The existing computing power can no longer keep up with user demand, and Nvidia's GPU is becoming in short supply again. "Even the banned models in stock have been sold out."

Will the Jevons paradox fail in the field of AI in the future? This is the reason why investors have different expectations for Nvidia. In the long run, whether the demand for computing power and chips will continue to increase depends on whether the increase in demand brought about by the expansion of AI applications can exceed the rate of decline in computing power costs. But now, AI competition has been incorporated into the narrative of great power games. When technological competition is mixed with political factors, everything will become more complicated and has deviated from the scope of simple market logic.

According to previous media data, the capital expenditures of several US technology giants have reached historical peaks in the past year, and more than half of them may have been "burned" on computing power. Although the market has been questioning the return on its computing power expenditure, the vigilance against China's AI and the rise of DeepSeek will still make the giants insist on spending money to pile up computing power for fear of falling behind. Google, Amazon and Meta's investment in computing power will continue to increase in the new year, and the US government is also vigorously promoting Stargate, an AI infrastructure plan that claims to be as high as $500 billion.

At least in the current situation, the rise of DeepSeek may not only fail to "kick Nvidia and slap Stargate in the face", but may instead give Nvidia an assist. Maybe even Trump may have to boast that his launch of Stargate was "foresighted".