What Everyone Gets Wrong About the AI Bubble

Markets may look chaotic right now, but history is pretty clear about what happens when a new technology captures global attention: capital pours in, valuations stretch belief, and everyone argues about whether it’s a bubble. What gets overlooked is the boring part — that these periods usually build the infrastructure of the next economy. AI is no exception.

AI appears to be in the third phase today. Sam Altman, OpenAI’s CEO, put it bluntly: parts of the field are “absolutely boiling,” yet “something real is happening.” Jeff Bezos calls it an industrial bubble with “massive societal benefits.” And he’s right. Overinvestment is rarely efficient, but it is often transformative. The dot-com excesses gave us cheap global bandwidth, hyperscale computing, and the foundations of today’s internet economy.

At Fintual we ran a simulation to illustrate this dynamic. We modeled a US$3,000 investment split between Booking Holdings, Cisco, and Amazon during the dot-com boom. The basket fell nearly 90% after 2000. Yet, by holding until 2025, it still delivered a 12.39% annualized return and ended near US$60,000. That’s the quiet power of technological cycles: they destroy illusions, not fundamentals.

Last month another signal appeared. OpenAI confirmed it plans to go public between 2026 and 2027. The company is already valued near US$500 billion privately, and public markets may assign it a US$1 trillion valuation. It expects to raise at least US$60 billion — capital that will help fund an unprecedented US$1.4 trillion infrastructure buildout with partners like Oracle and Nvidia. The numbers are staggering, and so are the losses: roughly US$20 billion in revenue, but near US$8 billion in cash flow, plus billions more in Microsoft-absorbed expenses. A trillion-dollar valuation implies a price-to-sales ratio of about 50×. Sustainable? No one really knows.

History is generous to the winners and brutal to the rest. Google multiplied its value almost 90× after going public. Amazon grew nearly 6,000× from its IPO despite losing money for years. Meanwhile, WeWork never made it to market, and Snapchat struggled for a decade. That asymmetry — limited downside, unlimited upside — is why investors chase early-stage opportunities, and why the 2025 IPO market is running hot again, with new listings up 41% on average, far ahead of the S&P 500.

But focusing on stock prices misses the real story. What matters is the infrastructure being built under the surface: chips, data centers, power systems, model-training networks, global compute supply chains. Few are watching that; most are watching the tickers. Yet, the infrastructure is what survives the crash and powers the next decade. This was true in the dot-com era, and it’s likely true now.

For investors, the implication is straightforward. The timing of AI corrections is unknowable, but the direction of long-term infrastructure investment is unmistakable. That favors diversified, long-horizon portfolios. Indexes tend to adapt, shedding companies that shrink and incorporating those that scale, capturing the economic residue of each technological wave.

While waiting for OpenAI’s IPO, the simplest way to gain indirect exposure is through its largest public shareholders — Microsoft and SoftBank — or through broad global equity indexes that will incorporate the eventual winners automatically.

Technological booms always feel exaggerated when you’re standing in them. But what they leave behind often becomes the platform for the next economic leap. The enthusiasm around AI may prove excessive in the short run, but the infrastructure being built today is likely to define the next era — and reward the investors patient enough to stay diversified while the cycle plays out.

If anything, the bigger risk for investors is misunderstanding what this phase represents. The market isn’t simply pricing OpenAI or Nvidia or the “Magnificent 7,” it’s pricing the possibility of a new technological substrate, much like electrification or mobile computing. That kind of shift rarely follows a clean trajectory. It swings between euphoria and panic, and only later becomes obvious that the real investment thesis was the compounding effects of infrastructure: cheaper compute, better models, and broader adoption.

In that sense, AI resembles a public-goods project disguised as a private-sector boom — the returns show up unevenly, slowly at first, and then all at once when new applications emerge. Consider how long it took cloud computing to become an unquestioned utility, or how fiber-optic overcapacity became the backbone of streaming, e-commerce, and remote work. We’re still early in AI’s equivalent of that timeline. The gap between hype and reality looks uncomfortable now, but it’s in that gap where future productivity gains usually incubate. Investors who wait for perfect clarity almost always arrive late. Those who allocate steadily, accept volatility as the cost of innovation, and diversify across geographies and sectors tend to capture the long-term payoff.