Lessons from Three Tech Bubble Bursts: How to Escape the "AI Bubble"?

Over the past two months, there has been a growing concern in the market about a so-called "AI bubble."

A few days ago, Polymarket, the world’s largest prediction platform, has launched a betting market on “When will the AI bubble burst?” The indicators they listed include a 50% plunge in Nvidia’s stock price, OpenAI or Anthropic declaring bankruptcy or being acquired, and semiconductor ETFs dropping by 40%. The results showed that 15% of participants believe the AI bubble will burst by next March, while 40% think it will happen by the end of next year.

This shows that many people have shifted from being optimistic to more cautious about AI’s future.

Concerns about the prospects of AI are based on a number of signs.

First, valuations of tech giants involved in AI have soared too rapidly.

According to research by CICC, since ChatGPT ignited the AI wave in 2022, the so-called “Magnificent Seven” in the US stock market—including Microsoft, Nvidia, Amazon, and Google—have risen by as much as 283%, far outpacing the 69% gain in the S&P 500 excluding these companies over the same period.

The same is true in China. Companies like Alibaba, Tencent, SMIC, Horizon Robotics, and Cambricon have all seen impressive gains.

Second, AI infrastructure spending is unleashing a “spending frenzy,” which could lead to overinvestment.

UBS statistics show that AI-related spending is expected to reach $375 billion this year, and exceed $500 billion next year. However, current revenues in the AI sector are nowhere near matching such huge investments.

Researchers have calculated that the current AI industry’s capital expenditure to revenue ratio is 6:1, much higher than that of previous tech bubbles (2:1 for the railway bubble, and 4:1 for the dot-com bubble).

Third, AI companies are playing a “left hand to right hand” game among themselves.

AI companies are investing in each other, effectively creating an internal revenue loop. For example, Nvidia invested $2 billion in xAI, which in turn borrowed $12.5 billion to buy Nvidia chips; Microsoft invested $13 billion in OpenAI, which then committed to spending $50 billion on Microsoft’s cloud services, and Microsoft subsequently purchased $100 billion worth of Nvidia chips.

This "supplier financing" model is blurring the boundaries between customers, suppliers, and investors—a scenario that also emerged during the internet bubble era. For instance, Cisco once financed its customers (telecom operators) to purchase its own equipment, enabling large-scale construction of fiber optic networks, which created artificial demand that resulted in massive idle capacity and eventually triggered a collapse of the entire industrial chain.

As a result, many people now believe that the current "AI bubble" is strikingly similar to the internet bubble of the late 1990s. Even industry leaders such as Microsoft founder Bill Gates, Google CEO Sundar Pichai, and OpenAI CEO Sam Altman have acknowledged the existence of a bubble and have compared its potential risks to those of the internet bubble.

So, how should we view the "AI bubble"?

Lessons from Three Burst Tech Bubbles

"Bubble" itself is actually a neutral term; it isn’t inherently negative. Every wave of technological innovation inevitably comes with bursts of investment and entrepreneurial enthusiasm. In fact, it’s often these seemingly "wildly imaginative" impulses that drive industrial progress forward.

However, when more and more people get involved and become overly optimistic about technological innovation, and as small-scale experimentation scales up to mass adoption, things can easily become disconnected from reality. This allows the bubble to inflate unchecked and eventually morph into a systemic risk that can no longer be avoided—concealing deep-seated dangers.

Therefore, the key is to recognize what stage the bubble has reached and how to identify the risks hidden within.

Mark Twain once said, “History doesn’t repeat itself, but it often rhymes.” This is especially evident in technological revolutions: every major disruptive innovation almost inevitably comes with a bubble and a subsequent crash. While the technology itself may achieve great success, behind the scenes, countless investors see their assets vanish.

The steam age had its own canal and railway bubbles.

After Britain launched its industrial revolution in the 18th century, factories and mechanized production began to emerge, but trains had not yet been invented. During this period, cargo shipping was handled primarily by barges, prompting a frenzy of canal construction in Britain. Vast amounts of capital and manpower were invested, and canal stock prices soared. However, canals required lengthy operational cycles to recoup investments, making them high-risk undertakings. In the end, railways did not give canals the time to succeed, gradually displacing them and ultimately causing canal stock prices to collapse.

Similar to canal investments, railways were touted as the “lifeblood” of the new industrial era after their emergence. Massive amounts of capital poured into the British railway industry, and many even leveraged themselves to invest. As a result, rail lines were laid everywhere—even in remote, impoverished areas with no matching commercial capabilities to support them. The industry’s overall rate of return kept declining, with actual revenues amounting to only a quarter of what builders had expected. This made it impossible to support the dramatically inflated stock prices. On top of that, a shift from loose to tight monetary policy abruptly brought this speculative frenzy to a halt.

In the electrical era, investment booms in aviation, automobiles, and electric infrastructure also triggered localized crises. A considerable number of investors suffered heavy losses in these technological revolutions because the path from the emergence of a technology to wide-scale adoption requires long and uncertain exploration of both technological direction and business models. Most people found it much easier to fall into traps than to seize genuine opportunities.

Take the automobile industry, for example. The dominant technological path was unsettled at first, with three main power solutions emerging: improved steam engines, electric motors, and gasoline engines. While we now know that gasoline cars ultimately prevailed, at the time, the more mature steam technology and the seemingly more promising electric motors occupied most of the market share.

Now, even if you had correctly identified gasoline-powered vehicles as the future and focused your investments on companies adopting that technology, the car industry was soon thrust into a phase of cutthroat competition. At its peak, there were hundreds of automotive manufacturers in the U.S., with over half surviving less than six years.

Even if you managed to spot then-fledgling industry titans like General Motors and Ford among so many companies, you’d still have needed to invest at the right time and sell at the right moment. General Motors alone nearly went bankrupt twice in the early 20th century—many investors gave up too early and missed its eventual resurgence. However, during the Great Depression, even General Motors and other car stocks experienced a sharp crash; those who missed the window to sell got caught out yet again.

The information age gave rise to an unprecedented bubble, which burst around the year 2000. Afterward, it took more than a decade for most stock markets to finally recover from the shadows of those losses.

Why was the damage so severe this time?

A major reason is that, during the internet bubble, the practice of disregarding dividends and cash flow as core indicators of company performance was taken to the extreme.

Investors were overly optimistic about the disruptive power of the internet, firmly believing that it would completely overturn traditional business models. Looking back, the internet certainly has been disruptive, but back then, viable business models had yet to emerge. Netscape, which made money selling browser licenses, and Yahoo!, a web portal, were seen as benchmark companies, and their stock prices soared by several multiples.

At the time, traditional valuation methods were cast aside. The yardstick for valuation gradually shifted from profitability to revenue, and then from revenue to concepts like website traffic and projected revenues for the next few years.

As a result, even some fledgling companies with no established business could command sky-high valuations. Sometimes, all it took was a business plan with an “e-” at the front and a “.com” at the end to secure tens of millions of dollars in real investment.

Of course, excessive optimism about new internet technologies was only one key driver of the bubble. Other factors contributed as well.

For example, at the time, interest rates in the US were quite low, and hot money from the Asian financial crisis flowed into the market, making liquidity abundant.

There was also a psychological aspect: as Robert Shiller pointed out in Irrational Exuberance, by the late 1990s, the internet had already entered people’s homes, providing richer entertainment and information than ever before. It wasn’t just investors who were extremely optimistic; ordinary people were as well. Many attributed economic growth from the recovery to the internet, this dazzling new star, believing it really was the engine of new economic growth—even though most internet companies at the time weren’t making much profit.

Additionally, the U.S. economy was the lone bright spot globally: the Soviet Union had dissolved, Japan’s economy was in the doldrums, Asia was facing a financial crisis, but the US economy was growing. All of these were seen as huge positives for the US stock market. Investing in US stocks was, for many people, as foolproof as investing in real estate had seemed a few years earlier: a surefire way to make money with no risk of loss.

New internet technologies and the U.S. stock market together attracted massive attention, sparking a booming expansion. Internet stocks became incredibly sought after, with Wall Street continuously launching IPO projects to satisfy the public's demand for internet shares.

The stock market bubble didn’t seem to sizzle—it burst all at once. On March 10, 2000, the NASDAQ hit a record high. Just days later, major internet companies released their financial results, which largely fell short of expectations, leaving investors feeling uneasy. At that time, Barron’s published a report pointing out that among more than 200 surveyed internet companies, 71% were unprofitable, and 51 were expected to run out of cash within a year. The article was titled “Burn Rate.”

The Federal Reserve also sensed the bubble had become too large and began raising interest rates. Some of the once high-flying internet stars that lacked sustainable business models were then delisted. The September 11, 2001, terrorist attacks dealt yet another heavy blow to the U.S. stock market. By October 2002, the NASDAQ had plunged to just a quarter of its peak value.

Stages and Identification of a Bubble

The bursting of technology bubbles can actually be seen as a result of the disconnect between financial capital and industrial capital.

Industrial capital is used to buy raw materials, equipment, and pay workers—directly engaging in production and the circulation of goods. Financial capital, on the other hand, enters the productive sphere indirectly, through investment, lending, or trading in securities.

Industrial capital operates in the real world: it buys machinery, develops technology, hires and trains talent, develops markets, and manages after-sales service, among other things. Financial capital, meanwhile, is focused on the virtual—making money from money by discovering and investing in lucrative projects. This means returns from industrial capital take a longer cycle to realize, while those from financial capital come much quicker.

From this perspective, it’s clear that the extent of harm caused by the three major tech bubbles depended on how far financial capital had diverged from industrial capital.

For example, during the electrical era, the bubbles created in industries like automobiles and aviation were relatively contained compared to the railway or internet bubbles, largely because financial capital hadn’t become completely disconnected from industrial capital. This is somewhat like the current new energy vehicle industry: while dozens of brands compete and there is some degree of bubble, it’s seen more as the result of internal competition and over-saturation rather than a financial system gone wild.

However, both the railway bubble and the internet bubble reached the extreme stage where financial capital, entirely divorced from cash flows, became the dominant force. At that point, financiers believed they could create value through speculative actions alone, treating industrial capital merely as an object to be manipulated and exploited. This led to an increasing disconnect between paper wealth and real wealth, between genuine profits or dividends and capital gains—until, inevitably, the whole system collapsed.

In her book Technological Revolutions and Financial Capital, British economist Carlota Perez outlines the relationship between financial capital and production capital at different stages. The phase where the two become highly divided is when the bubble inflates and bursts—a period also marked by the most intense disruption from technological revolutions. At the same time, however, new technologies take advantage of abundant funding to experiment wildly and lay down new infrastructure. Through short-term wealth creation, these new technologies and industries are widely adopted, making the use of new technology a common practice.

It is only after the bubble bursts, during the phase when financial and production capital work in tandem, that technology truly transforms industries. Financial capital returns to reality, constrained by new rules and institutions, and once again serves production capital. Most industries begin to apply the new technology on a large scale, as production capital resumes control, generating real growth and dividends.

This observation is also borne out in reality. For example, during the "railway bubble," stock prices soared around 1850 due to high expectations and then collapsed in 1857 amid panic. The genuine industrial dividends only became apparent in the early 1860s, once the national railway network was completed.

Similarly, after the burst of the internet bubble, true dividends were realized in the first decade of the 21st century as various industries began to transform.

Therefore, financial capital plays a role in absorbing uncertainty in advance, while production capital needs time to build infrastructure and reshape organizational forms. Ultimately, real wealth growth still depends on actual output.

From the perspective of the relationship between financial capital and production capital, it seems that the development of AI is still some distance away from the frenzy stage. For example, AI giants rely more on internal cash flow—that is, production capital—rather than on financial capital, with sizable and steadily growing revenue already achieved; currently, the "Magnificent Seven" stocks in the US market have a dynamic P/E ratio of about 30, while during the internet bubble, P/E ratios typically reached 50 or 60.

Moreover, although AI technology continues to advance, its disruptive impact hasn't yet become obvious in everyday life. Large language models often make errors, humanoid robots are widely ridiculed, and algorithmic improvements in AI have yet to deliver truly impressive breakthroughs.

This has yet to generate an "irrational consensus" on a mass scale like what we saw during the Internet bubble era. Back then, people were anchored in the industrial age and staring at a brand new information age ahead. The threshold for perceiving technological change was much lower, making the contrast more striking. In addition, there were unique international circumstances and shared psychological states influencing the times.

However, the fact that it hasn't happened yet doesn't mean it never will. History teaches us that disruptive innovation almost inevitably brings about bubble bursts—it's just a question of how big or small they are. So how can we avoid the damage caused by bubbles?

The book Bubbles and Crashes: The Boom and Bust of Technological Innovation points out a key pattern: it’s not technology itself that creates bubbles, but rather the resonance of three factors—high uncertainty, powerful storytelling, and tradable conduits—that triggers speculative frenzy.

• The Uncertainty Dimension

Uncertainty mainly stems from the technology path, market competition, business models and value chains, and market demand. The more uncertainty there is in these factors, the higher the potential for dramatic surges and plunges. Conversely, once these elements become clear, the likelihood of a bubble is greatly reduced.

• The Storytelling Dimension

“Before a company truly takes off, it’s really just a story about an ‘imagined future.’” In other words, every “technology narrative” reflects people’s various expectations and guesses about what the future holds—at its core, it’s a "bet" on the direction ahead.

The strength of a “technology narrative” partly depends on the scope for imagination. The more possible scenarios a technology can inspire about the future, the more likely it is to attract widespread attention and support.

Take the railway bubble, for instance. Railways were not just the arteries of bulk goods transport; they were also crucial for urban expansion, and, if extended to people's doorsteps, could revolutionize daily travel. They also promised to upgrade international trade significantly, thus drawing in investors from all walks of life. Similarly, the early days of the Internet inspired a vision of transformation for every aspect of life—food, clothing, housing, transportation—attracting people from different backgrounds to invest in this imagined future.

The other factor is the time required for realization. Narratives that are impossible to verify in the short term are much more likely to create bubbles. Building out infrastructures for canals, railways, electricity, the Internet, and even AI can take years or decades, and it is during this stretch that rampant speculation tends to flourish.

Finally, there is also the barrier of narrative comprehension. Technology is often complex, but if you can explain a complicated technology in a way that’s closely connected to everyday life, and the investment threshold is relatively low, it becomes much more shareable and influential.

• Tradeable Carriers

These are essentially companies that are tightly tied to a particular technology and carry its narrative. Investors see them as pioneers bringing the technology to life, and by buying their stocks, they're essentially investing in the technology narrative.

For example, Tesla has become the symbol of new energy vehicles, and Elon Musk is a public figure who constantly captures the public’s attention. Ultimately, Tesla delivered on the promises within its story. But that wave of new energy vehicle manufacturing wasn’t just about Tesla—names like Fisker, Dyson, Bright, and Coda also flashed by at lightning speed. They, too, carried this narrative and attracted some investors.

Generally speaking, when there are too many of these “carriers” and most of them lack real business performance, their valuations can only be driven up by storytelling and hype, which inflates the bubble further and further. This is similar to the dot-com bubble, when more than 70% of the 200+ internet companies were losing money—an extremely dangerous sign.

If you break down the above three points, you’ll actually be able to identify bubbles to some extent. For example, you can check the key variables related to uncertainty: whether there’s been a technological breakthrough, whether there are substantial production numbers, whether a competitive moat is being built, and how solid the profit model is. It’s also important to ask if the core narrative of a technology or company is still intact—is it gradually collapsing, or becoming even stronger? And how do the real business numbers of participating companies look? Is there a risk of cash flow drying up?

While it seems to come down to just three short points, the reality is much more complex. So, it’s important to acknowledge your own ignorance: don’t rush into investments you don’t understand, and never blindly follow the crowd.