Summary of Key Points
Tang Haiqing, the director of the Tianfeng Securities Research Institute, clearly stated at the Tsinghua Forum that there is no bubble in AI and presented his arguments from three perspectives: demand, valuation, and a productivity revolution. He also pointed out that this year is the best year for AI investment due to AI generating revenue on a large scale for the first time. To identify potential risks, one should consider three indicators: the Scaling Law, the cost-effectiveness of investments, and the supply-demand relationship. Finally, he recommended paying attention to sectors such as AI computing power (including optical modules), general-purpose PCBs/CCLs/MLCCs, and semiconductors.
1. Is There No Bubble in AI? Three Reasons to Convince You
Tang Haiqing believes there is definitely no bubble in AI, based on three solid arguments:
1. Demand is truly skyrocketing: Unlike the internet bubble in 2000 (when no one was using it), AI hardware is now in high demand—for example, NV series chips, which were available in quantities of 3 million a year ago but are now available in stock in quantities of 10 million, and all purchasing companies are making full use of the computing power. The demand is genuine.
2. Prices have risen significantly, but valuations remain low: Stock prices of AI-related companies have increased sharply, yet their actual valuations are still lower than those of consumer or real estate sectors. For instance, leading storage companies have seen hundreds of percent growth in their financial reports, while the top three global optical module companies have a PE ratio (a valuation indicator) of only around ten times next year.
3. This is a productivity revolution: AI computing power is analogous to steel during the Industrial Revolution; demand for it exceeded expectations back then, and the same is true now with optical modules. In 2023, 2 million 800G optical modules were sold in the first year, rising to 10 million in 2024, 20 million in 2025, and an expected 80 million this year (including new 1.6T specifications). The transition from one generation of products to the next is happening simultaneously, indicating a revolution rather than a bubble.
2. Why Is This Year the Best Year for AI Investment?
Tang Haiqing believes that AI has finally started to generate profits, similar to how new energy vehicles became profitable in July 2020:
- No revenue before, but now there is a significant breakthrough: After three years of development, AI has been capital-intensive. This year, companies like OpenAI and Anthropic have annual revenues exceeding $100 billion, with projections for over $250 billion next year, indicating a fully established business model.
- Revenue-generating areas are expanding: Currently, AI generates revenue mainly through education (e.g., AI-powered tutoring), customer service (AI robots), and programming (AI-generated code). In the future, this will expand into more industries, with long-term upward trends despite potential fluctuations.
3. How to Identify AI Risks in Advance? Look at Three Indicators
Tang Haiqing advises focusing on three fundamental principles. As long as these remain unchanged, there is no bubble in AI:
1. Scaling Law: This principle underlies the progress of AI—simply put, larger models and more data lead to better performance. Although some claimed that DeepSeek broke this rule, it was actually due to its strength in specific mathematical tasks rather than overall capabilities; leading companies like Anthropic’s Claud still rely on this principle.
2. Cost-Effectiveness of Investments: Each upgrade of an AI model (e.g., from ChatGPT4 to 5) requires five times the previous investment. With ChatGPT5 being considered a “college-level” achievement, there is room for further improvements in subsequent versions (postgraduate and doctoral levels), indicating ongoing investment needs.
3. Supply-Demand Relationship: In the next two years, AI hardware such as GPUs, optical modules, and fibers will be in short supply, with demand far exceeding supply, ensuring stable prices and demand.
4. Which Areas Are Worth Paying Attention To? These Offer the Best Value
Tang Haiqing recommended three key areas:
1. AI Computing Power (related to “optics”): This includes optical modules, fibers, and storage hardware. China leads the world in these fields—accounting for 60% of global fiber optic cable production and possessing advanced optical module technology. Currently, one GPU requires 5-6 optical modules (up from 2-3 previously), and this demand is growing faster than that for GPUs.
2. General-Purpose PCBs/CCLs/MLCCs: These are electronic components. Rising costs of upstream materials in the first half of the year will be offset by cost reductions in the second half, creating profit opportunities. Additionally, more overseas market share is shifting to China.
3. Semiconductors (storage and autonomy): In terms of storage, companies like Micron and SK Hynix are expanding slowly, while Chinese firms like ChangXin and Yangtze Memory are increasing production significantly. There is also a strong demand for domestic alternatives in semiconductor manufacturing.
5. Will Model Companies Face Losses and Withdraw from the Market? Maybe in the Future, But Not Yet
Regarding Professor Li Daokui’s question about whether model companies will withdraw due to chip costs, Tang Haiqing noted:
- Elimination is inevitable: However, we are still in a competitive phase where leading companies like Anthropic, OpenAI, and Google are investing heavily to seize opportunities. Although Meta has experienced some fluctuations, it has not fallen behind. Elon Musk’s investments are cautious but not abandoned. We are still in the middle of the process, not at the point of elimination.
Overall, Tang Haiqing is very optimistic about AI; he believes there is no bubble and that now is a good time to invest, particularly in hard technology areas such as computing power and semiconductors. For laypeople, this can be understood as AI being like the early stages of an industrial revolution—now is the time to buy related hardware, similar to investing in steel stocks back then, with significant opportunities but the need to monitor key indicators closely.