AI Bubble or Revolution? Investment Warnings from Tulip Mania to Tokens

When AI Meets the Ghost of Historical Bubbles

Whenever a technology sparks mass enthusiasm, historians and financial analysts invariably reach for the same textbook case — the Dutch Tulip Mania of 1637. Now, as the AI wave sweeps through global capital markets, this ancient analogy is once again on the table. From "tulips" to "tokens," humanity seems to perpetually cycle through the same excitement and anxiety.

The core question is: Are we witnessing an epoch-defining technological revolution, or yet another market bubble inflated by capital-driven sentiment? The answer is likely not black and white.

Historical Lessons from Tulip Mania

What Was Tulip Mania?

In 17th-century Holland, a single rare tulip bulb could fetch a price equivalent to a luxurious canal-side mansion in Amsterdam. People weren't buying tulips because they actually needed them — they were convinced that "someone will always be willing to pay a higher price" — the classic "Greater Fool Theory." The moment the last buyer disappeared, the market collapsed spectacularly, leaving countless speculators financially ruined.

To understand the deeper mechanics of this mania, we need to consider the historical context. The 1630s Netherlands was in the midst of its so-called "Golden Age" — thanks to its global trade network and financial innovations, the Dutch Republic was the wealthiest nation in Europe. The Amsterdam Stock Exchange had already been operating for over thirty years, and financial instruments like futures contracts were quite sophisticated. It was precisely this well-developed financial infrastructure that allowed people to trade without actually possessing the bulbs — what they were buying and selling was "the right to delivery next season," essentially a form of futures speculation. When the market suddenly crashed in February 1637, numerous contracts couldn't be fulfilled. Interestingly, however, research by modern economic historians (such as Peter Garber) suggests that the crisis's impact on the Dutch real economy may have been far less catastrophic than later literary works portrayed. Nevertheless, its symbolic significance as a "speculative bubble" has become deeply embedded in financial culture.

The "Greater Fool Theory," an important concept in behavioral finance, was formulated by economists analyzing speculative behavior. Its core assumption is: even when investors know an asset is severely overvalued, as long as they believe a "greater fool" exists who will buy at an even higher price, speculation will continue. This aligns with the "feedback loop" mechanism described by Nobel laureate Robert Shiller in Irrational Exuberance — rising prices attract more buyers, more buyers further push up prices, until this self-reinforcing cycle collapses due to some triggering event.

Common Characteristics of Asset Bubbles

Every asset bubble in history — whether tulips, the South Sea Company, the dot-com bubble, or cryptocurrency — displays strikingly similar patterns:

• Narrative-driven: A grand and alluring future story replaces rational cash flow assessment
• Valuation decoupling: Asset prices become severely detached from the actual value they generate
• FOMO sentiment: Fear of Missing Out becomes the primary buying motivation
• Leverage amplification: Capital floods in, further inflating prices and risk

Economist Hyman Minsky systematically described this process in five stages: displacement, boom, euphoria, profit-taking, and panic. In the "displacement" stage, a genuine innovation or policy change alters economic prospects; by the "euphoria" stage, speculators heavily leverage themselves chasing prices that have departed from fundamentals; ultimately in the "panic" stage, everyone tries to exit simultaneously, and liquidity evaporates instantly. This model can be almost perfectly applied to every financial crisis from the 1720 South Sea Bubble to the 2008 subprime mortgage crisis.

Examining "Token" (whether referring to the computational unit of AI large models or cryptocurrency tokens) within this framework, one can indeed identify quite a few alarming similarities.

It's worth specifically noting that the word "Token" carries a dual meaning in current tech discourse. In the AI technology domain, a Token is the basic computational unit through which large language models process text — a piece of text is broken down into Tokens before being fed into the model for processing. OpenAI's GPT-4 and similar models charge users based on Token consumption, making Tokens a real unit of value measurement in the AI economy, similar to "compute hours" in the cloud computing era. In the cryptocurrency and Web3 space, Tokens refer to digital coins issued on blockchains, representing certain rights or governance powers, with their value largely depending on market consensus and ecosystem development. What both types of Tokens share in common is: they are both imbued with narratives of "future value," and whether their current market pricing is reasonable often depends on which stage of the Minsky cycle you're observing from.

Is AI a Bubble or a Genuine Technological Revolution?

Key Differences from Pure Speculative Assets

Simply equating AI with tulips may seriously underestimate the substance of this technological transformation. Unlike flowers that hold only ornamental value, current AI technology is already tangibly changing productivity:

• Large language models are reshaping software development, content creation, and customer service workflows
• Enterprise AI applications are delivering measurable efficiency gains and cost savings
• Investments in underlying computing power and infrastructure are creating real value across the industrial chain

Specifically, the current AI industry chain can be understood across three layers of real value: The compute layer — the AI chip market led by NVIDIA GPUs experienced explosive growth during 2023-2024, with global data center capital expenditure expected to exceed $200 billion in 2024, creating genuine infrastructure value; The model layer — foundation models developed by OpenAI, Anthropic, Google DeepMind, and others represent billions of dollars in R&D investment and hard-to-replicate technical moats; The application layer — from GitHub Copilot providing code assistance to programmers (reportedly improving coding efficiency by 55% according to GitHub) to enterprise customer service automation reducing handling costs by 60-80%, these are real returns verifiable in financial statements. McKinsey's 2023 research report estimates that generative AI could create $2.6 trillion to $4.4 trillion in value for the global economy annually — a figure equivalent to the UK's entire annual GDP.

In other words, tulips never improved anyone's work efficiency, while AI is already generating actual economic returns. This is the most fundamental difference between the current boom and a pure speculative bubble.

Bubbles and Real Value Can Coexist

A more mature perspective holds that a technology's real value and market speculation bubbles can absolutely exist simultaneously. The dot-com bubble burst in 2000, but the internet itself fundamentally transformed human society. The investors who got burned were real, and giants like Amazon and Google that rose from the ashes were equally real.

The history of the dot-com bubble provides an extremely valuable reference. In March 2000, the Nasdaq Composite Index began plummeting after breaching its peak of 5,048 points, falling to 1,114 by October 2002 — a decline exceeding 78%. In this collapse, hundreds of "internet concept companies" like Pets.com, Webvan, and Kozmo.com vanished entirely — they burned through billions of investor dollars without ever establishing sustainable business models. However, during the same period, although Amazon's stock price fell from $107 to $7 (a 93% decline), the company survived and grew over the following two decades into a trillion-dollar behemoth. Google was founded in 1998 during the height of bubble mania, yet rose against the tide after the bubble burst thanks to search advertising as a killer business model. The core lesson from this history is: bubbles destroy the valuation bubble itself, not the value of the underlying technology. The fiber optic networks and data centers laid during the bubble era precisely provided the foundation for the subsequent explosion of mobile internet and cloud computing.

AI is very likely replaying this script: in the short term, some companies' valuations may be irrationally inflated and face severe corrections once market sentiment reverses; but in the long term, companies that truly master core technologies and find sustainable business models will become the winners of the new era. The AI space is already showing clear "valuation divergence" — on one hand, enterprises with large-scale user bases and mature revenue models (such as Microsoft, which has already achieved billions of dollars in incremental revenue through Azure AI services) are being reasonably priced by the market; on the other hand, a large number of startups that have secured hundreds of millions in valuation based solely on an AI-wrapped concept — a significant proportion of which may become this cycle's "Pets.com."

How Investors Can Make Rational Decisions Amid the AI Boom

Identifying Real Gold Within the Bubble

For investors and practitioners navigating this wave, the key lies in developing discernment:

1. Focus on cash flow, not narratives: Is an AI company truly solving problems and generating revenue, or merely surviving on concept-driven fundraising? When evaluating AI companies, focus on several core metrics — Annual Recurring Revenue (ARR) growth rate and quality, Net Revenue Retention (NRR, measuring whether existing customers continue expanding usage), and gross margin (reflecting technical moats and economies of scale). A healthy AI company should demonstrate a "T2D3" growth curve (revenue tripling for two consecutive years, then doubling for three consecutive years), rather than relying solely on one-off funding news to sustain its valuation.

2. Beware of valuations detached from fundamentals: When a team without a mature product receives an astronomical valuation, history's alarm bells should ring.

3. Understand technology's real boundaries: Know what AI can and cannot do, and avoid being misled by exaggerated claims. Current limitations of large language models include: hallucination problems (models confidently generating inaccurate information), reasoning capability ceilings, lag in accessing real-time information, and insufficient reliability in high-stakes decision-making scenarios. Understanding these boundaries helps assess whether a company's AI product claims are credible — if a product claims to achieve "100% accurate medical diagnosis" or "fully autonomous legal decision-making" using AI, this is almost certainly overhyped.

4. Diversify investments to reduce risk: Don't bet all your chips on a single track or a single company.

Maintaining Clear-Eyed Optimism

Facing the AI boom, the healthiest attitude is perhaps "clear-eyed optimism" — acknowledging the technology's revolutionary potential while remaining constantly vigilant against irrational exuberance in market sentiment. History doesn't repeat itself exactly, but it always rhymes.

The academic foundation for this attitude comes from Carlota Perez's "Technological Revolution Cycle Theory." In her classic work Technological Revolutions and Financial Capital, Perez argues that every major technological revolution (from the Industrial Revolution to the Information Age) goes through two major phases: first, a financial capital-dominated "installation period" accompanied by speculative frenzy and bubble bursts; then a production capital-dominated "deployment period" where technology is widely applied to the real economy, creating genuine prosperity. If this framework applies to AI, then we may currently be in the latter half of the "installation period" — the risk of a bubble is real, but it is precisely the necessary path through which new technologies are rapidly adopted by society.

Conclusion: Finding Certainty Amid the Frenzy

From Holland's tulip fields to today's data centers, humanity's fervor for new things has never changed. The difference is that tulips were ultimately just flowers, while AI may be the next general-purpose technology revolution after electricity and the internet.

Comparing AI to electricity is not casual rhetoric. Economists call such technologies "General Purpose Technologies" (GPT) — characterized by their ability to permeate virtually all economic sectors and continuously spawn innovation. Historically, fewer than twenty technologies have been recognized as general purpose, including the steam engine, electricity, the internal combustion engine, and the internet. The common pattern among these technologies is: from invention to fully transforming economic structures typically requires a 20-30 year "diffusion period," during which markets tend to first overestimate short-term impact (creating bubbles) and then underestimate long-term impact (creating buy-the-dip opportunities). If AI truly belongs to this category, its most profound impact on human economy and society may not fully manifest until the 2040s.

True wisdom lies not in predicting whether a bubble exists — it almost certainly does in some form — but in whether, after the bubble bursts, one can identify and seize the technologies and companies that truly change the world. When the tide goes out, we will ultimately see who was swimming naked, and we will also see who built solid levees.