Summary of Key Points
The current capital market is wildly betting on generative AI (GenAI), believing it will bring significant productivity improvements. However, historical experience shows that it takes several years or even decades for technological changes to translate into actual macroeconomic efficiency gains. The market is becoming "impatient" (this year's adjustments in the software sector are a sign of this), fearing that the short-term benefits may not be realized, which could impact chip demand, giant companies' investments, and the credit market. Experts warn that we need to pay attention to slowing sales growth in companies (a signal of a transition period) and closely monitor indicators such as credit gaps, debt levels, and data center vacancy rates. While the long-term trend of AI is clear, there could be sharp short-term corrections. As long as computing power is actually utilized, the risks are manageable.
Detailed Analysis
1. AI Productivity Isn't Instantaneous; History Tells Us to Wait
Does buying a new tool immediately increase efficiency? Not so simple. For example, electricity technology existed in the 1880s, but global efficiency didn't significantly improve until the 1920s-1940s (a delay of 40-60 years). Initially, factories simply replaced steam engines with electric motors without changing production lines, resulting in limited gains. It was only when each device used an electric motor and production lines were reorganized that efficiency truly soared. Tech companies like Amazon and Meta took 6-8 years to iterate on their infrastructure and business models before becoming industry leaders. Experts point out that it takes about 5 years for capital investment as a percentage of GDP to reflect in overall efficiency improvements. So, although AI is popular now, we still need to wait for its full impact on society.
2. AI Has Already Reaped Easy Benefits; Complex Areas Require More Time
AI can currently help with tasks like coding and text processing (which are relatively straightforward applications), which is why companies like Anthropic are seeing rapid sales growth. On the cost side, U.S. companies have seen a 2.2% year-on-year increase in efficiency (above historical averages), with per capita real income growing by 4.5% annually (four times the normal rate). The proportion of labor in total output has dropped below 53% for the first time since 1947, and corporate profit margins are near their highest levels. However, in complex fields such as healthcare and industrial manufacturing, AI has not yet had a significant impact, and further technological breakthroughs are needed to improve efficiency.
3. The Credit Market's Challenges: AI Is Costly, Making Borrowing Difficult
Building data centers and purchasing chips is very expensive, so companies that don't earn enough have to borrow money. Experts use the "financing gap" (the difference between expenses and earnings) as a measure of this pressure: a positive financing gap indicates a need for more borrowing. Historically, when this gap reached 2.8% of GDP, markets often peaked (as seen during the tech bubble in 2000 and before the 2022 stock market crash). Currently, the gap is decreasing, but we must be cautious. Additionally:
- Tech debt has surged: Cloud service provider debt in 2025 could be four times the average of the past five years, and global AI-related debt could reach $570 billion by 2026.
- Fewer people are buying bonds: The bond subscription ratio has dropped from 5 times to 2 times.
- Debt is being held off-balance sheet: For example, some of Meta's data center projects' debt is held by private investors, making it difficult for outsiders to assess the true risks. Fund managers are increasingly worried that these debts could trigger a credit crisis.
4. Short-Term Drops Are Possible, but the Long-Term Trend Remains Unchanged: Cyclical Corrections Are Normal
Even if AI has a long-term positive impact, there can still be sharp short-term declines. For instance, Amazon's stock price fell by 50% in 2022, and Nvidia's by 60%. An oversupply of data centers may lead to temporary setbacks, but whether this is a buying opportunity depends on whether AI can generate profits and whether companies can effectively utilize it. Experts say there's no need to panic: the current 1% vacancy rate in North American data centers indicates that computing power is being used, so there's no waste. A significant increase in vacancy rates (indicating wasted computing power) would be a real concern.
5. Monitor These Indicators to Assess AI Investment Risks
To assess the risks of AI investments, look at these key indicators:
- The frequency AI is mentioned in company financial reports (10-Ks) compared to per capita sales growth: If AI is frequently discussed but sales don't increase, it suggests that productivity gains have not materialized.
- The proportion of financing gaps as a percentage of GDP: Be cautious if it approaches 2.8%.
- Data center vacancy rates: High vacancy rates indicate wasted computing power.
- Interest rates on tech debt: If they are much higher than those on government bonds, it suggests that the market sees high risks.
These indicators can help you gauge the potential risks of AI investments and avoid making mistakes.