Has the AI bubble burst?
On June 5th, the global capital market experienced a rare "bloodbath." On that day, the U.S. stock market tumbled significantly. The S&P 500 Index and the Dow Jones Industrial Average dropped by 2.64% and 1.35% respectively, while the Nasdaq Composite Index plunged by 4.18%, marking the largest single - day decline since April 2025. The semiconductor sector became the hardest - hit area. The Philadelphia Semiconductor Index fell by 10.3% in a single day, recording its worst single - day performance since March 2020. NVIDIA, Broadcom, and TSMC dropped by 6.2%, 7.92%, and 6.76% respectively. The declines of companies such as Intel, AMD, Micron, and Qualcomm all exceeded 10%. AI software companies and cloud platform companies also suffered heavy losses. Among them, Amazon fell by 3.08%, Meta by 5.54%, Microsoft by 2.65%, and Oracle even by 9.65%.
The stock markets in Japan and South Korea, which are closely related to the AI industrial chain, also declined significantly. Japan's Nikkei 225 Index fell by 1.31%. AI and semiconductor heavy - weight stocks such as Tokyo Electron, Advantest, and SoftBank, which had previously driven the index to new highs, became the core forces dragging the index down on that day. South Korea's KOSPI Index tumbled by 5.54%. SK Hynix and Samsung Electronics, which had continued to skyrocket not long ago, became important forces leading to the index decline.
Since the start of this round of AI market in the first half of 2025, the AI - related sectors in multiple major markets have shown an almost one - way upward trend. Encouraged by slogans such as "AI will completely change the world" and "Buying AI is buying the future," a large amount of funds flocked into the AI track. Not only did the valuations of companies with excellent performance continuously reach new highs, but the stock prices of some companies that just caught the AI hot - spot also rose significantly. From this perspective, this sharp decline undoubtedly poured cold water on the fanatical market atmosphere.
Some commentators believe that the AI boom has caused the market to accumulate too many bubbles, and this "Black Friday" on June 5th may become the starting point for the bursting of the AI bubble. So, what exactly are the reasons for this sharp decline? Does it really mark the bursting of the AI bubble? What problems are hidden behind the concerns about the AI bubble?
What Caused the Sharp Decline?
The deep correction in the capital market is often not caused by a single factor. The sharp decline of global AI - related assets on June 5th was also the result of a series of factors working together.
First, the upcoming listing of giants such as SpaceX has triggered market concerns about the diversion of funds by large - scale IPOs. The sharp decline on June 5th occurred after SpaceX launched its global roadshow and before the official pricing. According to the documents disclosed by the U.S. Securities and Exchange Commission, SpaceX plans to issue approximately 555.6 million shares in this IPO, raising $75 billion, with a corresponding valuation of approximately $1.75 trillion. For investors, SpaceX, with its dual narratives of "space + AI," the endorsement of Elon Musk, and the halo of "the world's largest IPO," is undoubtedly a highly attractive target. To participate in the subscription, some investors need to prepare cash in advance. However, in the previous AI frenzy, many investors had already invested a large amount of funds in AI or semiconductor stocks and did not have sufficient cash. To obtain cash, they may choose to reduce their holdings of previously highly profitable stocks, thus putting pressure on the relevant stock prices.
There is more than one such large - scale IPO. Recently, Anthropic has secretly submitted an IPO application, and OpenAI is also reported to be preparing for listing, with their potential valuations possibly reaching the trillion - dollar level. These IPOs are like pumps that may divert a large amount of market liquidity, thus curbing the upward trend of the market.
Second, Broadcom's announced revenue guidance for AI chips fell short of expectations, leading the market to doubt the narrative of "unlimited explosion of AI computing power." After the U.S. stock market closed on June 3rd, Broadcom released its financial report for the second quarter of fiscal year 2026 and subsequent performance guidance. The revenue from AI semiconductors in the second quarter reached $10.8 billion, a year - on - year increase of 143%. However, the company expects the revenue from AI semiconductors in the third fiscal quarter to be $16 billion, lower than market expectations; the annual sales of AI semiconductors are expected to reach $56 billion, also lower than the expectations of some institutions before.
Broadcom is one of the core suppliers of global AI custom chips and computing power interconnection infrastructure, deeply covering cloud providers and AI giants such as Google, Meta, and OpenAI. Its orders and performance guidance are usually regarded as a barometer of global AI industry capital expenditure. Therefore, its revenue guidance falling short of expectations is interpreted as a marginal slowdown in the growth rate of AI industry capital expenditure. Even if this slowdown is just a normal change under a high base, it is enough to shake the market's optimistic expectations for computing power demand.
Third, the U.S. non - farm payroll data far exceeded expectations, triggering market concerns about the tightening of monetary policy. The number of new non - farm payroll jobs in the United States in May increased by 172,000, far exceeding market expectations; the unemployment rate remained at 4.3%, and wages increased by 3.4% year - on - year. The employment data for March and April were revised upward by a total of 93,000, indicating that the resilience of the labor market far exceeded previous forecasts.
Normally, strong economic data should be good news for the stock market. However, in an environment where the market is highly sensitive to interest rates, it has become bad news. Due to the robust employment market, the market's expectations for the Fed to cut interest rates have been weakened. Some institutions even believe that the Fed may raise interest rates. For technology stocks with high valuations and highly dependent on long - term cash - flow pricing, an increase in the discount rate will severely squeeze the valuations. Therefore, the situation of the stock market falling despite the good economic data has occurred.
The Root Cause Behind the Sharp Decline
The listing of SpaceX, Broadcom's revenue guidance falling short of expectations, and the U.S. non - farm payroll data exceeding expectations are just the direct fuses that ignited the sharp decline. Behind these superficial factors, this sharp decline also reveals deep - seated problems in AI development.
After the sharp decline, many commentators compared it with the bursting of the Internet bubble at the beginning of this century. This comparison does make some sense. The backgrounds of the two major adjustments in technology stocks are very similar: both were born in an environment of loose monetary policy and abundant liquidity; both relied on the grand narrative of the "next - generation productivity revolution" and experienced a super - bull market that deviated from the fundamentals to some extent; both witnessed a high concentration of funds, over - valued future growth, and capital speculation of concepts while the actual implementation lagged behind; and both occurred after the expectation of tightening monetary policy and the inflection point of liquidity.
This is not a coincidence. Behind the two sharp rises and corrections, there is a similar cycle logic. To understand this logic, the business cycle theory of the Austrian School may be a good auxiliary tool.
The business cycle theory of the Austrian School was systematically proposed by Ludwig von Mises and supplemented and improved by Friedrich Hayek. The capital theory of "round - about production" can be traced back to Eugen Böhm - Bawerk. According to this theory, economic growth comes from the deepening of the "round - about production" degree in the production structure. In order to improve long - term production efficiency, the economy will give up immediate consumption and invest resources to build longer and more complex intermediate production chains. The longer the chain, the greater the capital investment, and the greater the potential long - term growth space, but the risk of misallocation will also increase, and the economy will be more sensitive to interest - rate changes.
If the interest rate is determined by market supply and demand, it should reflect the social savings rate and consumers' time preferences. However, when the central bank continuously pursues a loose policy and artificially lowers the credit cost, the interest - rate signal may malfunction. Capital will misjudge future demand and over - flow into long - cycle, high - expectation, and capital - intensive tracks, resulting in over - investment, over - expansion, and structural deformities. Although this will bring short - term prosperity, it is essentially a false expansion driven by liquidity and an over - draft of future growth.
Once the monetary policy turns to tightening, the over - investment, over - valued assets, and excess production capacity previously supported by cheap credit will be exposed intensively. The market will spontaneously correct through falling asset prices, investment contraction, shutdowns, and layoffs, and the prosperity previously spawned by loose credit will also be liquidated.
Using this theory to observe this round of AI market, the logic becomes clear. The main reason for this round of market is of course the breakthrough in generative AI technology, but the relatively loose liquidity environment maintained by major economies from the end of 2024 to 2025 is also an important external factor. At that time, the decline in inflation and the long - term low interest rates almost became the mainstream market expectations. Many famous economists, including Olivier Blanchard, even published monographs specifically studying long - term low interest rates. It was under this expectation that a large amount of funds flocked into the capital - intensive and long - chain AI track.
The ultimate output of the AI industry is the intelligent efficiency improvement of enterprises, terminal intelligent applications, and productivity upgrades. However, before the final output is realized, the market needs to lay a long chain of intermediate capital goods. Upstream computing power hardware such as optical modules, storage media, and GPUs, and mid - stream computing power center infrastructure, large - model training, and algorithm iteration are all pre - conditions for downstream AI application development and industry implementation.
This industrial form of "first build equipment, then pile up computing power, then train models, and finally develop applications" determines that when liquidity is relatively abundant, funds will first flow into the upstream of the industry. Companies related to optical modules, high - end storage, AI chips, and computing power servers have become the biggest beneficiaries, and their valuations have risen again and again. In contrast, the commercialization of AI software and applications has not been so smooth. Although the valuations of related companies have also increased, the increase is relatively small. As for industries that are less related to the AI industrial chain, it is difficult to obtain the same amount of incremental funds, and the valuations of related companies are naturally difficult to rise. The stock prices of some companies have even continued to decline.
There is a popular saying in the market that buying the stocks of optical module companies means "standing in the light," while not buying means "the light is just there." Although it is just a joke, it vividly describes the phenomenon of funds concentrating on the upstream of the AI industrial chain.
However, a highly round - about production structure may cause serious misallocation. When the market pre - lays out the production capacity of the entire industrial chain based on long - term optimistic expectations, and cloud providers and technology companies hoard GPUs and reserve storage resources on a large scale, the improvement of AI production capacity may far exceed the real downstream demand. If this contradiction continues to accumulate, the misallocation of funds will deepen. Once the interest - rate trend changes, the entire chain may collapse from the upstream. From this perspective, the sharp decline on June 5th can be regarded as a self - correction of the market for the misallocation of funds. It shows that the industrial structure round - about and valuation bubbles caused by loose credit have accumulated to a certain extent and need to be released intensively.
Is the Bubble About to Burst?
So, does this sharp decline mean that this round of AI market has reached its peak and the bubble is about to burst completely? At least so far, the situation doesn't seem so pessimistic.
First, from the perspective of fundamentals, the current stock - market bubble is still at a relatively controllable level. During the Internet bubble in 2000, the market was full of shell companies with no revenue, no profit, and no implementation scenarios. The entire industry had almost no positive cash - flow support and relied entirely on capital "burning money" to survive. In contrast, the major listed companies in the current AI track all have relatively stable cash flows. Needless to say about upstream hardware companies such as NVIDIA. Although AI platform companies such as Microsoft and Google cannot yet obtain profits matching their investments from AI business, their other businesses can still provide sufficient cash flows. Even if they cannot obtain direct returns from AI investments for the time being, they have the ability to continue to bear R & D and infrastructure expenditures through their mature businesses.
Second, from the perspective of valuation, the price - to - earnings ratios of major companies are also at a relatively moderate level. NVIDIA's current price - to - earnings ratio is about 31.8 times, while those of Microsoft, Google, and Amazon are 24.5 times, 27.7 times, and 29.3 times respectively. On the eve of the bursting of the Internet bubble, Cisco's price - to - earnings ratio was as high as more than 220 times, Microsoft's was more than 60 times, and Amazon had not yet achieved profitability. Although the valuations of current AI and semiconductor companies have increased significantly, this increase is largely due to the improvement in revenue and profits. As long as the profits of major enterprises can continue to grow, there is a possibility that the valuations can be gradually digested through profit growth.
Third, in the short term, it is unlikely that the Fed will raise interest rates. One of the risks that the market is most worried about is that the Fed will restart interest - rate hikes. But in my opinion, the possibility is not high in the short term.
On the one hand, the economic performance reflected in the U.S. non - farm payroll data is not as good as it seems. Although the number of new jobs is impressive, the new jobs are mainly concentrated in sectors such as leisure and hospitality, local government, and healthcare. Some service - sector jobs may be related to the centralized recruitment before the "World Cup." Therefore, a single - month data is not enough to prove that the U.S. economy has reached the point where interest - rate hikes are needed to prevent overheating.
On the other hand, part of the current inflation pressure in the United States comes from supply factors such as the Middle East war, and interest - rate hikes cannot effectively contain it. If the war ends, the relevant inflation pressure may quickly decline. Although Jerome Powell was nominated by Donald Trump as the Fed Chairman, monetary policy still needs to be decided collectively by the Federal Open Market Committee, and the possibility of a hasty interest - rate hike is low.
Finally, both China and the United States currently regard AI as a key national development area and have invested a large amount of funds for support. Considering the relevant industrial policies, even if there are already some bubbles in the market, it is unlikely to burst in the short term.
Overall, although the sharp decline on June 5th has sounded the alarm for the bubble in the AI field, the bubble is far from bursting. Enterprises lacking performance and with over - valued stocks may face greater pressure, but enterprises with sufficient profit support may still experience a long - term high - growth period. If people can solve the structural problems in AI development during this period, the market may gradually compress the bubble and smoothly return to the path of healthy development.
Three Important Issues
So, in the current AI boom, what important structural problems need to be solved first? In my opinion, there are three issues that deserve the most attention.
The first issue is that the rapid technological iteration has led to the idling of cutting - edge technologies, while the implementation of applications has seriously lagged behind.
In the past three years, AI technology has been iterating rapidly. The capabilities of large models, computing power performance, multi - modal technology, and intelligent agents have been continuously breaking through. AI can already achieve many functions such as content generation, intelligent decision - making, and automated production, and seems to be on the verge of being transformed into actual productivity. It is under such a technological vision that the capital market has given extremely high valuation premiums to AI enterprises.
However, when it comes to the real - world industry, the technological glory has fallen into the dilemma of "idling in implementation." Many cutting - edge AI technologies still remain at the laboratory demonstration and small - scale pilot stages and are difficult to penetrate into actual industrial scenarios. The improvement of technological capabilities does not mean that enterprises already have the corresponding data foundation, organizational processes, and willingness to pay. After many models enter specific industries, they still need to solve problems such as accuracy, stability, responsibility division, and system access.
The important reason is that the supporting system cannot keep up with the technological iteration: industry standards are imperfect, data - security barriers are high, the digital foundation of traditional industries is weak, the cost of scenario adaptation is too high, and the compliance and regulatory system is also imperfect. Under the influence of multiple bottlenecks, there is still a long conversion chain between model capabilities and industrial productivity. The progress of model capabilities will not automatically be converted into corporate revenue and social productivity.
The lag in AI implementation will also lead to wasteful in - house competition in the industry. Many enterprises invest huge amounts of money in piling up computing power and training models. An important reason is that it is difficult to break through in the application end. In order to maintain high valuations, enterprises can only return to pure technological competition. If such in - house competition continues, it will not only waste resources but also make the false boom in the AI industry more intense. Enterprises continuously launch new models with larger parameters and higher training costs, but do not create enough paid demand synchronously. The resulting