Summary of Key Points
The AI computing power industry currently has a clear bubble, but experienced practitioners and investors are hesitant to go short. The reason is that the negatives are evident (giant companies facing tight cash flows, U.S. restrictions on computing power leasing, and the emergence of alternative models in China), yet the positives always come unexpectedly (such as ChatGPT, Claude, and the sudden surge in demand for computing power driven by applications like “Lobster”). No one dares to bet that the next positive factor won’t emerge. The article concludes by advising investors to be rational and not to be blinded by an obsession with either only the positives or only the negatives.
Three Clear Negatives in the Computing Power Industry:
1. Giant Companies Running Out of Cash: Foreign giants like Google and Meta, as well as domestic companies like Alibaba, may appear to have decent financial results, but the cost of purchasing computing power is spread over several years. In fact, they are spending much more cash than is reflected in their financial reports. For example, Alibaba recently issued additional shares to raise funds due to insufficient cash flow for computing power purchases. If this continues, they will either have to reduce their purchases or continue to issue debt or shares, which is unsustainable.
2. U.S. Restrictions on Computing Power in Southeast Asia: Many computing power centers in Southeast Asia are actually operated by Chinese companies, which account for a significant portion of NVIDIA’s global market share (second only to the U.S.). The U.S. is investigating and restricting Chinese companies from leasing computing power in the region. If these restrictions are strictly enforced, NVIDIA’s performance could plummet, as it would lose a large market.
3. Chinese Open-Source Models Competing with Commercial Models: The growth of Anthropic (the parent company of Claude) has slowed, partly due to competition from OpenAI, but more importantly, Chinese open-source models are becoming more viable alternatives. These models require less computing power and can be deployed privately, ensuring data security. As a result, many European and American companies are turning to Chinese open-source models to save costs. This suggests that the era of reckless expansion in the computing power industry may be coming to an end.
Why No One Dares to Go Short?
Despite the clear negatives, no one bets that the bubble will burst because past positives have always been unexpected:
- The Rise of ChatGPT: No one expected AI chatbots to become so popular, dramatically increasing global demand for computing power.
- Claude’s Success in Programming: Anthropic’s programming capabilities have created a new revenue stream for computing power, making it more profitable.
- The Rise of Domestic Chips: The emergence of domestic large models has revitalized the domestic computing power and chip markets.
- The Success of Agent and “Lobster” Applications: The sudden popularity of applications like Agent last year and “Lobster” this year has pushed global demand for computing power to new heights. These unexpected developments have prevented the bubble from bursting.
How Unexpected Positives “Prolong the Bubble”?
Every time people think the bubble is about to burst, some unexpected positive factor emerges:
- In 2023, many believed the industry’s rapid expansion would lead to a bubble burst, but the sudden popularity of an AI application (e.g., “Lobster”) caused companies to buy more computing power.
- Before ChatGPT, the computing power market was relatively calm, but its success led to a surge in demand for NVIDIA’s chips.
These unexpected breakthroughs create new demand, turning the market from a cooling trend into a hot one.
Investment Advice: Don’t Be Obsessed with Either Positives or Negatives
The article uses two examples to warn investors:
- Investors in ZeroRun and Duolingo focused only on the positives of their stocks (such as sales and user growth) and ignored other risks (policy changes, increased competition).
- The online education industry saw many positives (rapid user growth, significant funding) before a sudden policy change caused the entire industry to decline.
For the computing power industry, even if you see all the negatives now, don’t be overly pessimistic (what if there’s a “Lobster 2.0”?) Conversely, even if you see many positives, don’t be overly optimistic (what if the U.S. strictly restricts computing power in Southeast Asia?). Investors should accept that both positives and negatives coexist and be prepared for unexpected reversals.
In Conclusion: The computing power industry is in a situation where negatives are clear but no one dares to go short, and positives are unknown but no one dares to bet that they won’t happen. Understanding this logic helps you understand the market and develop a more rational investment approach—don’t focus solely on one aspect and be prepared for uncertainty.