Summary of Key Points
After a year and a half of rapid growth, the global AI industry is collectively "deflating its bubble": The market values of AI-related stocks both domestically (such as Zhipu in China and MiniMax) and internationally (including SpaceX and NVIDIA) have significantly shrunk, with some companies experiencing a more than 50% reduction in their market value. The root causes of the bubble include excessively high valuations ("market dream multiples"), limited practical applications (with only AI programming being relatively mature), and a decline in demand for computing power. The industry is shifting from a phase of "burning money to create narratives" to one of seeking commercialization models. While the long-term trend remains positive, a short-term return to rationality is necessary.
1. Immediate Signs of the Bubble Bursting: Global AI Stocks Tumbling
AI-related stocks have seen sharp declines recently:
- Domestically: Zhipu's stock price has plummeted from a peak of HK$2,980 to HK$1,107 (a market value reduction of over HK$48 billion), and MiniMax’s has fallen from HK$1,330 to HK$216 (a remaining market value of HK$75.4 billion), both experiencing a 50% decrease in market value.
- Internationally: SpaceX's stock price rose to $225 within three days of its listing but later fell below its issue price (resulting in a market value loss of $1.24 trillion); NVIDIA’s market value has shrunk by 15% ($900 billion), Microsoft’s by 31%, and Oracle’s by 40%.
- Indices and Markets: The NASDAQ 100 index has retreated by nearly 6%; the South Korean stock market has suffered even more, with its index soaring to 9,400 points in the first half of the year due to AI-related storage concepts before crashing to 6,800 points. Samsung and SK Hynix have each lost 31% and 38%, leaving investors trapped at the "peak."
These declines mark the official start of the AI industry's deflation process.
2. The Core of the Bubble: Excessively High Valuations
The most prominent feature of the AI bubble is the use of "market dream multiples" rather than actual performance as the basis for valuations:
- The Myth of Anthropic: Currently the fastest-growing AI company, it reported annual recurring revenue (ARR) of $47 billion in May this year (compared to only $9 billion at the end of last year), with internal forecasts anticipating cash flows of $17 billion by 2028. Investors have driven its valuation from $380 billion in February to $1.2 trillion in July.
- Other Companies Falling Short: OpenAI’s annual recurring revenue for this year is only $25 billion, and it may not become profitable until 2030; SpaceX's AI business is expected to generate $4.5 billion in revenue by 2026 (a growth of only 41%); although Zhipu’s annual recurring revenue has increased to $1 billion, it cannot support a market value in the tens of billions.
Investors have applied the high growth rate of Anthropic to all AI companies, leading to significantly inflated valuations for most of them.
3. Challenges in Practical Applications
The future prospects for AI are promising, but there are very few profitable use cases at present:
- Only AI Programming is Mature: While AI can indeed improve efficiency by helping programmers write code (for example, with the widespread adoption of the OpenClaw framework, companies like Alibaba and Tencent have seen their R&D staff using 200–500 million tokens per day), the number of users utilizing AI programming tools is limited (out of a total of 5 million programmers in China). Among ChatGPT’s 1 billion monthly active users, only 8 million use the Codex tool, indicating limited potential for expansion.
- Other Applications Are Not Practical: Examples like using AI to complete tasks (such as shopping or booking taxis) are not widely adopted by the general public. For instance, using AI for shopping may not save much time; in fact, it might require additional confirmation steps, making it less efficient. Other applications such as AI-assisted decision-making and collaborative work are still not well-established, and their commercial potential is far smaller than that of programming.
The limited number of practical use cases has led to a slowdown in the growth rate of token consumption. The quarterly growth rate of cloud service providers’ businesses has dropped from 13% to 9%, and token prices have also fallen by nearly 20% (from $2.8 per million tokens to $2.31).
4. Industry Adjustment Signals: Moving from "Burning Money for Scale" to "Finding Profitable Models"
As the bubble bursts, AI companies are starting to take more proactive measures:
- Decline in Demand for Computing Power: Microsoft has terminated a $3 billion computing power contract with Oracle, Blackstone has abandoned a $100 billion data center project, and SpaceX is renting out its computing resources to companies like Anthropic (rather than investing heavily in expanding its own capacity). The shift from a situation where there was a shortage of computing power to one where there is excess indicates a slowdown in industry growth.
- Capital Shifts to Profitability: Previously, AI companies focused on acquiring users through investment; now investors are more concerned about whether their AI businesses can generate profits. Many companies are seeking funding and preparing for IPOs to build cash reserves for the post-bubble era.
- Giants Reducing Spending: The five major tech giants in Silicon Valley have invested over $700 billion this year, but as stock prices have adjusted, the logic of "investing to increase market value" no longer holds. They are now focusing on controlling their expenses and identifying profitable use cases.
5. Future Trends: No Complete Collapse, but a Return to Rationality
The AI industry will not repeat the mistakes of the internet bubble:
- Stronger Resilience of Players: The industry is led by cash-rich giants like Microsoft and Google, as well as star companies backed by private equity (PE), which are more resilient to risks than the startups of the internet era.
- High Technical Barriers: AI requires substantial technical expertise, making it difficult for companies to defeat their competitors simply by spending money, thus avoiding a zero-sum game.
- Unchanged Long-Term Trends: According to Masayoshi Son, the impact of AI could be 10–50 times that of the internet boom. The current deflation is just a temporary adjustment, and the industry will move from excessive hype to more practical applications that align valuations with actual performance.
In summary, the deflation of the AI bubble is not a bad thing; it is a necessary step for the industry to mature from its current state of frenzy.