Summary of the Key Points
This article discusses the current state of the AI bubble in Silicon Valley: the bubble is expanding (IPO valuations of giants like SpaceX and OpenAI are absurdly high, with the amount of capital raised being ten times that of last year's U.S. stock market IPOs). However, industry participants—founders, investors, and engineers—are aware of the risks but no one is leaving the scene. Instead, they continue to bet on it while quietly preparing for alternatives (such as signing short-term contracts, seeking acquisition opportunities, or choosing less consensus-driven technical approaches). The bubble has also led to new evaluations of AI companies' valuations, international expansion strategies, and technological pathways. The article compares the differences between Silicon Valley, which tolerates bubbles, and domestic markets, which are more focused on practical profit-making.
I. The AI Bubble is Not Just “Abstract Figures”; It’s a Real “Game of Postures” and “Short-Term Contracts”
The bubble isn’t about abstract GDP-level investments; it’s about concrete realities:
- For large companies: Investing in AI is a matter of not wanting to fall behind. Many large firms do so because it’s part of their CEO’s annual plans, and not investing means admitting lagging behind. A MIT report states that companies have spent tens of billions on generative AI, with 95% of projects showing no returns, yet they continue to invest—this is about buying a sense of security rather than value.
- For startups: They use short-term contracts to create growth curves. Companies rarely sign three-year agreements; instead, they offer six-month trial periods. Startups include these contracts in their funding pitches, presenting them as signs of rapid growth, without anyone really caring if the contracts are renewed. Both upstream and downstream parties inflate this narrative: large companies use short-term contracts as a signal of interest, startups use them to raise funds, and investors resell them—no one calls this out.
- For newcomers: They are swept up in the need to tell compelling stories. Early entrants have high valuations, and latercomers, unable to match their performance, must also create attractive stories to secure funding. The bar for customers is getting higher, forcing newcomers into the bubble.
Of course, there are real examples of success, such as ElevenLabs’ voice AI technology, which led large companies to shut down their customer service centers and sign six-to-seven-year contracts—but such tangible achievements are rare in Silicon Valley.
II. Those Who Stay Have Their Own “Betting Strategies” and “Lifeboats”
Everyone sees the bubble, but no one leaves because everyone has their own calculations:
- Founders: Some bet on scale, while others bet on patience.
- Aggressive players, like Genspark (which raised $60 million in its seed round with no revenue), saw annual recurring revenue (ARR) exceed ten million nine days after launching its product and started advertising extensively. They bet that their large scale will make them viable or attractive to acquisition by larger companies—top-tier unicorns are looking for acquisition opportunities at high valuations as a backup plan.
- More cautious players, like Xiangfeng (a former executive from Baidu and Xiaomi), have tried three AI projects and are now focusing on market research. They bet that once the bubble deflates, their stable businesses will survive—preferably by targeting real unmet customer needs.
- Investors: Some bet on talent and genuine demand.
- Some break with traditional principles; for example, Holly invests in teams of top AI researchers even without a product, because in Silicon Valley, scarce combinations of talented individuals are key to successful projects.
- Others focus on customers; Bryan invests in AI companies with stable revenue (not dependent on the bubble), such as those serving traditional industries (e.g., replacing customer service).
- Engineers: They bet on irreplaceable value. Engineers observe internal shifts within companies—Alibaba is moving from open-source to closed-source, ByteDance is balancing commercialization with academic research, and Google is reorganizing its Gemini team. With ten years of experience becoming obsolete in new AI paradigms, what do they bet on? For example, Thomas argues that the only remaining value is the ability to take responsibility for outcomes: while AI can perform tasks, someone must be there to handle issues (like needing a driver in Tesla’s FSD).
III. Five “Clues to a Waking Up” Caused by the Bubble
The bubble has disrupted old rules, forcing industry participants to rethink:
1. Evaluating AI companies based on five criteria:
- Quality of revenue (is it real money? Is customer churn high?);
- Computational autonomy (are you not at the mercy of cloud providers/chip manufacturers?);
- Moats (do you have unique advantages that others cannot copy?);
- Capital efficiency (is the money invested where it’s needed?);
- Compliance (are you avoiding legal pitfalls?).
2. International expansion is a must: When registering a company, consider four questions: where are your core markets? Where does your technology/team come from? Where will you list in the future? Is the raised capital a resource or a liability?
3. Technological pathways are not set in stone: The current mainstream is transformers (the larger the model parameters, the better), but this path has physical limitations (computing power cannot be increased indefinitely). Alternative approaches, such as diffusion and non-autoregressive techniques, could prove more viable—don’t treat the current trend as the only correct one.
4. The business model for AI in B2B has changed: In the past, SaaS focused on attracting more users; now, with low marginal costs for AI, high-value individual customers are more valuable than a large user base. Growth is measured by how much money each customer generates, not by the number of users.
5. The AI industry needs to be stratified: For example, the AI video sector can be divided into four layers: model layer (monopolized by large companies), tool layer (with tens of millions in ARR), distribution layer, and community layer. Which layer is hottest? If you can’t clearly define your position, you haven’t truly entered the industry.
IV. Silicon Valley vs. Domestic Markets: Different Sizes of Bubbles; Practicality Is Our “Shield”
Silicon Valley has a higher tolerance for future possibilities, leading to larger bubbles; domestic markets are more pragmatic, with everyone asking when they can make money. This pragmatism was once seen as a limitation but is actually safer in a bubble environment because domestic startups focus on real revenue and are less susceptible to inflated valuations.
When the bubble bursts, what’s washed away is the excess, leaving behind talent, experience, and these new insights. Only time will reveal who retains the real value and who just has air.
The core message of this article is that although the AI bubble is large, smart people don’t make foolish bets. They participate while preparing for the worst and continue to evolve within it. What happens in Silicon Valley today could be our future tomorrow; understanding these trends early can help us avoid pitfalls.