Hello! I'm your financial analysis assistant. This article, written by a Ph.D. in economics from Shanghai University of Finance and Economics, serves as a dose of “rational cold water” to the currently overheated AI stock market, while also providing a “long-term perspective.”
The author’s core argument is quite sharp: Current AI stocks are mainly being supported by the reckless spending of “upstream companies” (those that supply the hardware and infrastructure for AI), but the “downstream users” who actually utilize AI have not yet spent much money. In the short term, there is a severe mismatch between supply and demand, which poses significant risks. In the long run, whether AI can truly transform human productivity (and thus GDP) will be the key determinant of the ultimate fate of these stock prices.
Let me break down this complex article into five easy-to-understand points to help you fully grasp the logic behind it.
---
1. Why are AI stocks so expensive now? Because we’re using “past results” to predict “future stories”
The article starts by pointing out that people currently judge the value of AI stocks based on traditional indicators such as the price-earnings ratio (PE) and price-sales ratio (PS).
Simple explanation:
It’s like going to a newly opened internet-famous restaurant and the owner tells you, “Last year I made 1 million, and I expect to make 2 million this year, so I’m worth 10 million.”
- The problem is: Both profits and sales figures are results. During good economic times, profits are high, and stock prices are high; during bad times, profits plummet, but stock prices don’t drop as much (because they can’t become negative). This leads to an illogical phenomenon where the price-earnings ratio is often highest after a crisis (since the denominator (profits) has decreased, but the numerator (stock price) hasn’t changed much).
- The current situation: The current price-earnings ratio of the S&P 500 (28.7 times) is much higher than the historical average (16.9 times) and even higher than during the high-valuation period of 1990. Although some use “forward price-earnings ratios” or the “Shiller P/E ratio” to smooth out these fluctuations, you’re still using events that have already happened to predict **events that haven’t yet occurred.
- Conclusion: These indicators can tell you that things are more expensive now than in the past, but they can’t tell you how much AI really deserves to be valued. AI is not a simple cyclical industry; it’s a revolution, and you can’t use old metrics to discover new possibilities.
2. Upstream companies are making huge profits, but downstream companies are losing money quickly
This is one of the most interesting parts of the article. The author introduces the concept of capital expenditure (Capex), which refers to the money large companies (like Google, Microsoft, Amazon) spend on buying chips and building data centers for AI.
Simple explanation:
Think of the gold rush:
- Upstream companies (suppliers): NVIDIA, Micron, Broadcom, etc. These companies are receiving a flood of orders because big businesses are spending heavily on equipment, resulting in impressive cash flows. Data shows that the free cash flows of several AI infrastructure companies are expected to reach $430 billion in the next year, three times what they were two years ago. This is why chip stocks have been rising—because people are actually spending real money.
- Downstream companies (users): Google, Microsoft, Meta, etc. They’ve spent a lot on equipment, but they haven’t yet made a profit. Data indicates that the free cash flows of these five companies are expected to turn negative for the first time in 2024.
- The core contradiction: Most of the money made by upstream companies comes from the spending of downstream users. If downstream companies continue to lose money, how long can the upstream prosperity last? It’s like a family where the father (upstream) makes money by selling things, and the son (downstream) spends all the money without generating any income, eventually leading to financial problems. You can’t judge the health of the AI industry just by looking at the excitement on the upstream side; you need to see if the downstream companies can make a profit.
3. How much money have users actually spent?
Since downstream companies are losing money, are end-users (you and me) paying a lot? The author uses data to debunk this misconception.
Simple explanation:
- Low payment rates: 1.8 billion people worldwide use AI, but only 3% are willing to pay for it. The remaining 97% are essentially getting something for free.
- Total spending is tiny: Global spending on AI models and platforms in 2025 is only $39.3 billion. In contrast, companies are investing tens of billions of dollars each year. What users spend is a fraction of what companies spend on equipment.
- Potential growth:
- Based on computers: If AI becomes as widespread as computers (with a current penetration rate of about 22%), there’s still a lot of room for growth.
- Based on smartphones: If AI becomes as popular as smartphones (with a global penetration rate of 60%), the market size could nearly triple.
- Potential payment market: If we calculate based on the number of new computers bought in the next five years, the market size for paying AI users could be 26 times what it is now, amounting to about $1 trillion.
- Price trend: Don’t expect prices to keep rising. Historically, as technology becomes more widespread, prices tend to decrease (think about computers and smartphones). The current decline in the price of AI “tokens” (units of computing power) indicates that people are more concerned with value for money, not just reckless spending.
4. Short-term imbalances, long-term focus on GDP
The article concludes with a macroeconomic indicator: the Buffett Index (total stock market value/GDP).
Simple explanation:
- Why GDP matters: Profits can fluctuate, but GDP (a country’s total output) is relatively stable. If AI can truly improve efficiency, it should increase GDP.
- Current warning sign: The Buffett Index for the U.S. stock market has reached a record high (over 220%). This means that stock prices have risen too quickly compared to the actual economic growth.
- Two possible outcomes:
1. Short-term adjustment: Stock prices may fall back to a more reasonable level.
2. Long-term verification: If AI truly boosts productivity and drives significant GDP growth, it will increase the “denominator” and help absorb the high valuations.
- Current assessment: AI is indeed being applied in industries like manufacturing and healthcare, but it hasn’t yet had a widespread impact on the overall economic structure. Therefore, in the short term, insufficient demand is a major risk; in the long term, the value of AI depends on its ability to drive GDP growth.
5. Summary: A guide for ordinary investors
Putting these four points together, we can establish a clear logical chain:
1. Overvalued: Traditional valuation indicators suggest that U.S. stocks are too expensive, and they’re not entirely suitable for a revolutionary industry like AI.
2. Risk of a broken supply chain: Upstream companies are making huge profits, but downstream companies are losing money. If the downstream companies continue to struggle, the prosperity of upstream companies is unsustainable.
3. Weak demand: End-users are not willing to pay much, and total consumption is far from covering the massive investments of upstream companies. Although the long-term potential is huge, it’s still a future reality, not the present.
4. Macroeconomic divergence: Stock price gains far exceed economic fundamentals (GDP), indicating a potential for a return to average levels.
Advice for ordinary investors:
- Don’t chase high prices blindly: Current AI stocks, especially those in the upstream hardware sector, have already exceeded future expectations.
- Watch for the “turning point in cash flows: Pay attention to when companies like Google and Microsoft stop losing money and start making profits. Once their free cash flows turn positive, it will signal that AI is truly generating commercial value, and that would be a safer time to enter the market.
- Be cautious of declining token prices: If the cost of using AI continues to decrease, it indicates that the technology is becoming more widespread, which is good, but it may also squeeze the profit margins of upstream companies.
- Long-term perspective: If you believe AI can change the world like electricity or the internet, then short-term fluctuations are just noise. However, understand that realizing the value of AI is a long process that requires time to prove its ability to drive GDP growth, not just rely on hype.
In one sentence:
The story of AI is exciting, but the costs involved are substantial. Upstream companies are celebrating, downstream companies are struggling, and users are waiting and watching. Short-term risks outweigh opportunities, and the long-term value of AI depends on its ability to truly boost GDP, not just on its hype.