Deputy Chairman of the Federal Reserve: Analyzing the AI bubble from four perspectives

At a time when NVIDIA’s market value continues to reach new highs, and AI-related concepts are gaining popularity in the US stock market as well as the cryptocurrency sector, capital is showing increasing interest in “smart narratives.” From computing chips to AI-powered blockchain projects, valuations are rising simultaneously, which has raised concerns among investors about potential bubble risks. In 2025, Federal Reserve Vice Chairman Jefferson stated during a recent financial stability meeting that the current AI boom shares similarities with the Internet revolution of the 1990s. He identified key indicators for assessing whether the market is overheating, including valuation expansion, financing structures, profitability, and risk transmission. This discussion about whether this is a technological revolution or just another speculative cycle is becoming a crucial dividing line for global funds in revaluating risky assets.

This statement not only reflects the cautious approach of the world’s most important central banks towards emerging technologies, but also provides market participants with a clear framework for evaluating the AI boom.

I. The Federal Reserve’s Perspective: Dual Mission and Financial Stability

All policies and observations by the Federal Reserve are centered around its dual mission—maximizing employment and maintaining price stability. Jefferson emphasized that evaluating the impact of artificial intelligence must start from this fundamental task. This means that the Federal Reserve focuses on AI not just for its technological breakthroughs or market performance, but also for how it affects overall employment levels, labor productivity, economic growth potential, and inflation trends.

From an employment perspective, AI has both positive and negative effects. On the one hand, AI helps to promote employment by improving work efficiency and creating new jobs, such as positions related to AI development, deployment, and maintenance. On the other hand, its automation replacement effect may lead to the decline of certain occupations, especially affecting young workers with less experience. Jefferson pointed out that if AI only replaces existing labor forces without creating new jobs, it could cause short-term economic slowdown. This balance between replacement and complementation is crucial in assessing the structural impact of AI on the labor market.

From the perspective of price stability, AI’s ability to increase productivity helps reduce production costs, thereby putting downward pressure on prices. Applications such as efficient resource allocation, supply chain optimization, and decision-making assistance can help curb inflation. However, at the same time, the construction of AI infrastructure, such as data centers, can drive up the prices of inputs like land and energy. Additionally, the increase in salaries for AI professionals could lead to cost-push inflation. These two-sided effects make the net impact of AI on inflation highly uncertain, requiring continuous monitoring.

To fulfill this dual mission, a stable and resilient financial system is essential. The Federal Reserve continuously monitors systemic risks through its semi-annual Financial Stability Report (FSR). Recent surveys show that 30% of market participants consider “changes in attitudes toward AI” as a significant risk to the financial system, which is a significant increase from the 9% reported in spring. This suggests that if market expectations about AI suddenly change, it could lead to tighter financial conditions and economic downturns. Therefore, by including AI in its financial stability monitoring framework, the Federal Reserve aims to prevent potential asset bubbles and financial vulnerabilities caused by the rapid development of AI technologies.

II. Monitoring Framework: FSR and Market Sentiment Tracking

The Federal Reserve’s monitoring of AI is not conducted in isolation, but is integrated into its overall financial stability assessment framework. FSR focuses not only on traditional risks such as leverage ratios, asset valuation, and financing risks, but also takes into account the structural changes brought about by emerging technologies. Jefferson emphasized that policymakers must distinguish between “cyclical fluctuations” and “structural changes,” and AI likely falls into the latter category. This means that the productivity gains brought about by AI could alter the relationship between employment and inflation, thereby affecting the transmission mechanism of monetary policy.

Market sentiment is one of the key areas of focus for FSR. Surveys show that nearly one-third of market participants are aware of the potential risks associated with a reversal in market sentiment due to AI. This consensus could potentially become a “self-fulfilling prophecy” – once optimism shifts, rapid capital withdrawal could lead to significant adjustments in asset prices. Compared to the Internet bubble era, today’s information dissemination speed and widespread use of algorithmic trading may exacerbate market volatility. Therefore, the Federal Reserve’s monitoring of sentiment indicators serves as an early warning system for potential systemic risks.

Additionally, the application of AI in the financial industry itself presents new challenges for monitoring. AI tools such as high-frequency trading, robo-advisors, and risk models can improve efficiency, but they may also trigger new risks related to homogenization and pro-cyclical behavior. The Federal Reserve is working to strengthen its ability to identify and assess these emerging risks by expanding its analytical tools, including the use of AI technology itself.

III. Four Core Indicators: The Touchstone for Judging AI Bubbles

Jefferson identified four key differences between the current AI boom and the internet bubble in the late 1990s. These differences can serve as crucial indicators to determine whether there is a serious bubble in the AI industry today.

(1) Profitability: From “story-driven” to “profit-oriented” approaches. During the internet bubble, many companies went public based solely on the “.com” domain name. They lacked sustainable profit models, with minimal or even zero revenue, and had to rely on external financing and market hype to survive. In contrast, leading companies in the AI field today (such as some tech giants) have robust and diversified sources of income. They generate revenue not only through AI services but also by integrating AI deeply into their existing product portfolios, thereby enhancing the competitiveness of their core businesses. This “profit-oriented” approach gives AI investments more solid foundations, reducing the potential for pure speculative trading.

However, Jefferson also pointed out that the activity in private equity markets may partially mask the profitability challenges faced by early AI companies. Many venture capitals have poured into AI startups. Although these companies are not publicly traded, their high valuations mean they could still become risks if they fail to generate profits in the future. Therefore, it’s important to consider both public and private markets when evaluating profitability metrics.

(2) Valuation levels: Relative restraint in price-to-earnings ratios.

During the peak of the internet bubble, the price-earnings ratios of many internet companies reached hundreds or even thousands of times their actual value. This reflected the market's irrational optimism regarding future growth. Currently, although the stock prices of AI-related companies have risen significantly, their price-earnings ratios are still far below historical highs. This indicates that investors are still taking into account the companies' actual profitability and cash flows when investing in AI-related stocks.

Of course, the rationality of valuation should be determined by considering both industry characteristics and the stage of growth. As a versatile technology, AI has great potential for long-term value creation, and a moderate premium is justified. However, if valuations rise too quickly without considering fundamental factors, it could lead to bubbles. The Federal Reserve pays attention to valuation indicators in order to distinguish between rational enthusiasm and signs of overheating in the market.

(III) Number of listed companies: The scope of speculative activity is limited. In 1999-2000, more than 1,000 internet companies went public, creating a situation where any company could become popular simply by changing its domain name to “.com”. Currently, there are only about 50 listed companies explicitly classified as “AI core enterprises” (based on certain criteria), which is far fewer compared to the number of companies during the internet bubble. This suggests that speculative activities are relatively concentrated and have not spread across the entire market.

But Jefferson also warned that the private equity market may hide many AI startups. Although these companies are not publicly traded, they still engage in active fundraising activities. If these companies go public or their financing conditions change, they could become new sources of instability. Therefore, the indicator for “number of companies” needs to be monitored dynamically, including both public and private markets.

(IV) Financial leverage: The degree of reliance on debt is relatively low.

During the Internet bubble period, many companies relied on equity financing, with limited use of debt leverage. This helped to mitigate the direct impact of the bubble burst on the financial system. Currently, AI companies also tend to rely less on debt financing, which helps to limit the spread of risks. However, recent trends indicate that in order to support the massive investments in AI infrastructure such as data centers and computing clusters, some companies are beginning to increase their bond issuance and credit financing.

Jefferson specifically pointed out that as AI moves from software-level applications to hardware infrastructure, the demand for capital investment increases significantly, potentially leading to a gradual rise in leverage ratios. If sentiment around AI changes adversely, companies with high leverage will face greater debt pressure, and risks could spread to a wider range of economic sectors through credit channels. Therefore, it is important to closely monitor the trends in leverage indicators.

Fourth, these insights can be useful for market participants.

Jefferson’s analysis provides a valuable framework for policymakers, as well as important lessons for investors, businesses, and researchers:

First, when analyzing issues, it is necessary to start with the fundamental tasks of observers. Investors should go beyond short-term market sentiments and conduct in-depth analysis of the real impacts of AI technology on a company’s fundamentals, such as profitability, cost structure, and competitive barriers. Companies should focus on how AI can improve their productivity and long-term competitiveness, rather than blindly pursuing concepts.

Second, it is important to distinguish between cyclical fluctuations and structural changes.AI represents a technological revolution that could last for decades, and its impacts are structural in nature. In times of market fluctuations, it’s important to distinguish between long-term trends and short-term noise, avoiding misjudging structural opportunities as cyclical bubbles, or vice versa.

Thirdly, pay attention to the overall market response and systemic risks. The rise of individual companies or sectors doesn’t necessarily indicate a bubble. It’s necessary to assess the overall market valuation levels, capital concentration, leverage levels, and emotional trends.特别需要注意的是，AI-related narratives may shift from being driven by profitability to being driven by stories.

Fourthly, make use of analytical tools, including AI itself. AI technology can be used to more accurately assess market risks, corporate value, and economic impacts. Professionals should actively utilize data analysis and machine learning tools to improve decision-making quality, while also being aware of the potential new risks associated with model homogenization.

Fifthly, participate in a rational and enthusiastic manner, across multiple dimensions and dynamically.

Jefferson’s conclusion is relatively cautious and optimistic: based on comparisons across four dimensions—profitability, valuation levels, number of companies, and financial leverage—there are significant differences between the current AI boom and the internet bubble from the late 1990s. The likelihood of repeating such a dramatic collapse is low. AI development is rooted in a group of mature companies with stable profits, and the overall financial system is relatively resilient.

However, uncertainties still exist. The long-term impact of AI on employment, inflation, and productivity remains to be verified; market sentiment may reverse; the activity in private equity markets may mask risks; and there’s concern that infrastructure investments might increase leverage levels. Therefore, the Federal Reserve will continue to monitor the development of AI, ensuring that its growth takes place in a stable and resilient financial environment. Ultimately, this effort aims to serve the fundamental goals of maximizing employment and maintaining price stability.

For the market, Jefferson’s analysis provides a useful tool for rationally evaluating investments in AI. In the midst of technological revolutions and surges of enthusiasm for capital, it is important to remain clear-headed, distinguish between essence and appearance, and focus on long-term value. This may be the best way to avoid bubbles and embrace change. Is AI really a bubble? The answer lies not in simple binary choices, but rather in continuous, multi-dimensional, and dynamic observation and judgment.