When Duan Yongping started buying NVIDIA

Duan Yongping has started buying NVIDIA. Today, he said: “Although I don't fully understand, I'm preparing to sell a little put on NVDA. NVIDIA is actually a very good company.”

Last year, Teacher Duan commented on one of our articles backstage, pointing out a small mistake. That day’s significant drop in NVIDIA was due to the option expiration date. Teacher Duan should have been continuously observing NVIDIA, but the focus of his discussions seems to be on Apple and Tencent.

Starting to buy today means what? It means NVIDIA has reached a new stage.

1. From the beginning of the year to now, especially with the flourishing ecosystem activated by China's DS open source, the computing power structure has shifted from training to inference at an unexpected speed. The nature of computing power expenditure has also shifted from R&D investment to operational investment. The biggest characteristic of operational investment compared to R&D is that it is more closely linked to continuous business growth and has stronger relevance. In simple terms, the sustainability of computing power investment has strengthened. This relates to whether this business is ultimately a stable cash flow-generating good business.

(From another perspective, NVIDIA's 10-K disclosed its major direct customers, with three accounting for 11-12% of revenue, indicating a trend towards customer diversification.)

1. The biggest controversy or narrative is the shift from hardware to applications. If we try to view it from Teacher Duan's perspective, daring to draw a comparison, the mobile phone has also shifted from early hardware to a dazzling and prosperous application ecosystem, but today Apple remains the largest company in the world (3.6 trillion). The expansion of application value does not necessarily mean a decrease in hardware value because the entire cake is expanding. Referring to the development process of past computing platforms, applications and hardware should pull each other along, spiraling upwards together. It is possible that the value proportion of applications has indeed increased, but aside from the overall cake getting bigger, the competitive landscape of hardware, especially profit share, may be far better than that of applications (applications are likely more fragmented than hardware, but Apple takes 90% of the profits in the mobile industry). Even today, despite the GB200 delay and the unique Chinese market dominated by geopolitical forces, NV still hasn't seen a decent competitor.

At today's earnings meeting, management continued to mock ASIC: “ASIC design activity doesn't always translate into deployment.” ASICs often only see initial proposals, with few large-scale deployments. Indeed, the past waves of ASIC “cyclical speculation” essentially stem from the CSP's will always being a factor, but the reality is ultimately very honest.

For example, after 4 days of DS open source (which we had basically predicted in previous articles), we can see that DS is “extremely reliant on N cards” because for DeepSeek, which focuses on training, N cards are the most time- and resource-efficient tools, leaving no time to adapt or run alongside other computing power cards. On the other hand, in inference, domestic implementations have achieved 0 to 1, but in real production deployment environments, the ultimate competition is about price and ROI. Even the 4090 N card can demonstrate extremely high cost-effectivenessOf course, the biggest controversy analogous to Apple may be whether CUDA has the same ecological barriers and brand mindset as Apple's iOS (to C vs to B), and whether it can transition from earning hardware profits to earning software profits. These are likely to be observed over time, and cannot be completely denied or confirmed at this moment.

What I want to say is that the current market style, diversity of capital choices, narratives, etc., have spread from last year's infrastructure to applications, but NV's unique ecological position and actual value proportion are not necessarily weakened.

1. To be precise, NV's current price corresponds to a 20 times PE for 2026, making it one of the cheapest in M7. The low valuation reflects the current level of crowded positions, controversies over the long-term sustainability of the business, and various short-term noise disturbances. These may just align with the appetite of long-term value investors, as they can look through the short term, anchor a long-term point (but this point must be very solid), and use long-term strategies to pick up short-term bargains (yes, selling puts is just taking the chips from you hedge funds).

2. Finally, trying to empathize from a long-term perspective, how to continue participating in AI in the most "lazy" way. In fact, there are not many choices that can simultaneously meet long-term certainty and follow the linear growth of AI. The model layer has already started to be mixed up by DS open-source, and the application layer is actually hard to grasp before the model stabilizes. We have mentioned many times that Tencent and Apple, which firmly hold super entry points, can always capture the application value released by AI development. The demand for computing power is extremely certain, and while there has been narrative controversy in the competitive landscape, the fact has always been NV. Besides, what else can guarantee that it won't derail during the rapid development process in the next 5-10 years? For long-term investors, Apple, Tencent, and NVIDIA may be the most certain choices for participating in AI.

Of course, I might add one more point at the end. Recently, China's so-called "technological confidence" and America's "de-mystification of technology" essentially reflect that over the past three years, the whole world, including ourselves, has underestimated our engineering dividend and innovation capability. The financialization and hollowing out of manufacturing in the U.S. pose hidden dangers to its continued leadership in global innovation (Silicon Valley engineers are heavily invested in stocks, and the supply chain for robots and electric vehicles is severely dependent on China). Especially when an industry moves from 0 to 1 and begins 1-10, it relies heavily on engineering, which is China's advantage. Geopolitical division, on the other hand, forcibly carves out a monopolized market for Chinese enterprises. This may also be a huge variable in the future distribution of NV and the entire AI value. But this question seems to have no answer at the moment; let's observe it together.

Risk Warning and Disclaimer

The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial conditions, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstancesInvest based on this, and you bear the responsibility