How big is the current AI bubble? UBS: This round of bubble is more "reasonable" than during the TMT period, and the three major topping signals have not yet appeared

UBS believes that although the current U.S. stock market meets the seven prerequisites for a bubble cycle, the "rationality" of this round of AI bubble far exceeds that of the year 2000, and key peak events have yet to occur.

According to news from the trading desk, on October 29, UBS's global equity research team published a report indicating that we are in the early stages of a potential bubble, but the three key signals that indicate a bubble peak—extreme valuations, long-term overheating catalysts, and short-term peak events—have not yet appeared.

The report points out that the strong productivity enhancement potential of generative AI, along with the current higher risk of government balance sheets compared to corporations, provides a more solid foundation for this round of valuation expansion than during the 2000 internet bubble.

UBS found that the current market perceives a 20% probability of a bubble, and investors need to understand the key signals that indicate a bubble burst, which will be central to future investment decisions.

The "Rationality" of the Bubble: Why Is This Time Different?

In October, the Federal Reserve lowered interest rates as expected. UBS believes that the current U.S. stock market has met all seven prerequisites for a bubble: stocks have outperformed bonds by 14% annually over the past decade, significant new technologies have emerged, it has been 25 years since the last bubble, overall profits are under pressure, the market is highly concentrated, retail investors are buying, and monetary conditions are loose.

Nevertheless, UBS emphasizes that simply equating the current AI craze with historical bubbles is one-sided. The logic behind this round of bubble formation is more "rational" on two key levels compared to the internet bubble or the late 1980s Japanese bubble.

First, the disruptive potential of generative AI and its unprecedented adoption speed are unique.

The report points out that due to a large amount of infrastructure already in place, the adoption speed of generative AI far exceeds that of any previous technological revolution. For example, OpenAI attracted 800 million users in just three years, while it took Google nearly 13 years to reach a similar scale.

If the market expects generative AI to temporarily boost productivity growth by 2% like during the internet bubble, it would be enough to support a 20-25% upside in the stock market.

Second, the macro risk structure of this cycle has fundamentally changed.

During the 2000 internet bubble, the U.S. government enjoyed a budget surplus and had a very healthy balance sheet. Today, the government debt-to-GDP ratio is twice that of then, with high fiscal deficits. In contrast, the balance sheets of corporations (especially tech giants) are relatively robust.

(The condition of the U.S. government's balance sheet is much worse than that of corporations, and UBS believes this will lower the equity risk premium.)

UBS believes that this "weak government, strong corporations" pattern may lead investors to shift funds from nominal assets (such as bonds) to real assets (such as stocks) to avoid sovereign credit risk, thereby reducing the required equity risk premium (ERP) and supporting higher stock valuations.

No Sign of Bubble Peak One: Valuation Has Not Yet Reached Extremes

According to UBS's analysis, historical bubble peaks are usually accompanied by extreme valuations, but the valuations in the AI-related fields are far from dangerous levels at present.

First, absolute valuation levels still have distance to cover. In past bubbles, at least 30% of stocks by market value had price-to-earnings ratios rising to 45-73 times (based on 12-month trailing P/E ratios), and the yield on 10-year government bonds reached at least 5.5%.

Currently, the P/E ratio of Mag 6 (excluding Tesla) is only 35 times, well below bubble levels.

(P/E ratios of US stocks during bubble periods)

Second, from a relative valuation perspective, the premium on tech stocks is also within a normal range. Historical peaks typically occur when the Equity Risk Premium (ERP) drops to around 1% (as seen in 1929 and 2000).

According to UBS's calculations, the current ERP using its long-term model is still around 3%, indicating that the market has not completely ignored risks due to excessive optimism.

Additionally, the assumptions about the Total Addressable Market (TAM) are not "fantastical."

At the peak of the internet bubble, the valuation of telecom stocks required the assumption that households would spend about 20% of their income on telephone services, which was clearly unrealistic.

For the current AI-centric semiconductor industry, UBS estimates that if semiconductor industry spending reaches 1.3% of global GDP by 2030 (up from the current approximately 0.7%), then the current valuation is reasonable.

(Proportion of semiconductor industry vs. software industry in GDP)

Considering the importance of semiconductors and software as "the new oil," they currently account for about 3% of GDP combined. The historical average for oil was 3%, with peaks reaching 10%, making this assumption not far-fetched.

(Proportion of oil in US GDP)

Finally, the current investment logic has not deviated from fundamentals. During the Japanese bubble period, the market priced stocks based on land values. In the internet bubble period, "number of eyeballs" became the core metric. Now, profits and cash flow remain the main basis for analyzing AI-related companies.

No Sign of Bubble Peak Two: Lack of Long-term Overheating Catalysts

In addition to valuations, long-term structural factors that trigger bubble bursts, such as over-investment and excessive leverage, are also not evident at present.

Research reports emphasize that signals of over-investment have not yet appeared. The proportion of Information and Communication Technology (ICT) investment in GDP in the United States is still below the peak level of 2000, approaching normal levels. This indicates that a capital expenditure frenzy across the entire society in the United States has not yet formed (The proportion of technology investment to GDP is below the long-term average level)

The risk of excessive debt financing is also relatively low. Although the capital expenditure/sales ratio of the top 11 hyperscalers is approaching the level of telecom operators in the year 2000, the financing methods are completely different; today's tech giants mainly rely on their strong cash flow rather than debt to support investments.

UBS estimates that these companies need to increase their capital expenditures by 40% (based on 2025 revenue) before they will start to use debt financing. This stands in stark contrast to the net debt/EBITDA ratio of telecom companies during the internet bubble, which reached as high as 3.5 times.

(The ratio of capital expenditure to sales for large cloud computing companies)

The degree of market breadth narrowing is also not as severe as in 1999. At that time, the Nasdaq rose by 86%, but the number of declining stocks was nearly twice that of rising stocks, indicating extreme market differentiation. Currently, although market breadth has narrowed, such extreme divergence has not occurred.

(Comparison of the S&P index with the number of stocks rising/falling from 1998 to March 2000)

Finally, current U.S. corporate profits are also exceptionally robust. During the internet bubble, U.S. national accounts profits (NIPA) were actually declining, which explained the sharp deterioration in market breadth at that time. Currently, although profit growth is concentrated in a few leading companies, overall national accounts profits have not shown similar pressure.

Signal Three of No Bubble Peak: No Short-Term Peak Events

On a short-term level, the "starting gun" indicating that the market is about to peak has not yet been fired. UBS lists three potential peak-triggering events:

Giant M&A Transactions:

• The peak of the internet bubble was accompanied by century-defining mergers, such as Vodafone/Mannesmann and AOL/Time Warner.
• Adjusted for today's market size, these transactions amounted to approximately $900 billion. Currently, there has not been a similar level of M&A frenzy in the market.

(The proportion of M&A transaction amounts to S&P market value and the 6-month moving average of M&A transaction amounts)

Central Bank Policy Constitutes a Fatal Blow:

• Historical data shows that only when the Federal Reserve's policy interest rate exceeds the nominal GDP growth rate will it deliver a fatal blow to bubble formation.
• In 2000, the Nasdaq index doubled during the Federal Reserve's rate hike cycle until the policy interest rate rose above 6% (close to the nominal GDP growth rate at that time), after which the market peaked.
• Currently, the nominal GDP growth rate for 2026 is expected to be 5.2%, and monetary policy is far from tightening to the extent of "choking growth."

Weak earnings and extreme price momentum:

• During the internet bubble, earnings momentum peaked about a year before the market topped. Currently, the earnings revision momentum of tech stocks still outperforms the market.

(Earnings momentum peaked in the first quarter of 1999, with the red line representing the date of the market peak)

• At the same time, price momentum has not reached extremes; for example, the semiconductor sector's stock prices are currently about 35% above their 200-day moving average, while at the peak in 2000, this indicator reached 70%.

In summary, UBS's analysis provides investors with a detailed "bubble map." Despite the AI frenzy, key indicators from valuation, macro catalysts to short-term triggers suggest that this feast may be far from over.

However, UBS also warns that the real bubble may be lurking in the tech industry, especially in the high profit margins of the semiconductor sector, which may face pressure in the future as capital intensity increases and competition intensifies.

(Profit margins of tech stocks versus semiconductor stocks)