Decoding the AI Bubble: Tulips, Dot-Coms, or Something Else?

The following article was written by Shlomo Benartzi, professor emeritus of behavioral decision-making at UCLA’s Anderson School of Management and senior fellow at Wharton AI & Analytics Initiative, and Stefano Puntoni, Wharton marketing professor and co-director of Wharton Human-AI Research. 

The rise of artificial intelligence, and generative AI in particular, has triggered a familiar debate: Are we living through the early stages of a technological revolution — or yet another economic bubble? Depending on who you ask, AI is either the engine of the next century of productivity or the most overhyped technology of all time. These competing narratives shape how capital is allocated and how business leaders plan for the future.

This leads to a central question: If AI is a bubble, what kind of bubble is it?

History offers two classic analogies. The first is tulip mania, a frenzy driven almost entirely by speculation, where prices detached from any real underlying value. (At the height of tulip mania in the winter of 1636, a single rare tulip bulb sold for the price of a house in Amsterdam.)

It’s easy to laugh at tulip mania from a distance, but our era has produced its own versions. The cryptocurrency dogecoin, for example, increased in value by more than 200 times in the first few months of 2021, largely fueled by social media posts. The coin has since lost almost all of those gains. At least the tulip buyers received flowers for their trouble.

The second kind of bubble is the dot-com boom, where a genuine technological revolution collided with investor exuberance — producing short-term overvaluation but long-term transformation. Amazon’s share price, for instance, declined by 94% between late 1999 and late 2001, after the dot-com bubble popped. And while it took nearly ten years for the share price to bounce back, it is now 41 times higher than its dot-com peak.

Or consider Pets.com, which raised tens of millions of dollars in its IPO despite losing money on every bag of dog food it shipped. By the end of 2000 — less than a year after its celebrated public offering and Super Bowl commercial — the website had exhausted its resources and was forced into liquidation.

Crucially, however, the premise of Pets.com — that consumers would eventually purchase pet supplies online — proved prescient. By 2025, 55% of all pet food sales occurred through online channels. The failure of the firm was not evidence against the potential of e-commerce, but rather a consequence of timing, execution, and premature speculation.

Understanding whether AI resembles tulips or the internet is essential for interpreting today’s valuations and the expectations behind them. For executives allocating scarce capital, it is perhaps the most critical question of our time.

To help answer the question, let’s start with a simple decision tree (see Diagram 1).

Thinking Architecture for Evaluating Bubbles

Step 1: Is the Technology Providing Value to Enterprises?

Evaluating whether AI is in a bubble begins with a foundational question: Is the technology delivering meaningful value to firms?

Our three case studies (see below) suggest that AI is already creating significant, and potentially transformative, value. Of course, these success stories may be outliers. To better assess the overall value of AI, let’s look at a new study by Wharton researchers Jeremy Korst, Stefano Puntoni and Prasanna Tambe, along with a team at GBK. According to their survey of 800 enterprise executives, roughly 75% indicate that generative AI has already improved productivity, enhanced decision-making, or streamlined tasks. While these improvements often stem from early-stage implementations rather than fully embedded systems, they nonetheless represent meaningful benefits.

However, other research offers a different perspective.  A recent study from MIT, looked at organizations that have attempted to integrate AI into core processes, redesign workflows, and measure the financial impact of the technology. The researchers apply a stringent definition of success, and focus on whether or not AI has already led to profit and loss (P&L) level gains. Under these criteria, only 5% of firms have achieved success with AI.

While these two studies appear to be contradictory, a closer look demonstrates that they are seeking to answer fundamentally different questions.

Wharton asks: Is AI useful today?

And the answer appears to be yes. Many organizations are benefiting from generative AI in tangible ways — streamlining tasks, enhancing productivity, and improving decision-making.

MIT asks: Is AI transforming businesses already?

And the answer so far is not yet. Only a small fraction of firms have the governance, the infrastructure, and redesigned workflows required to convert experimental value into durable financial returns.

Taken together, the studies indicate that AI is providing real value, but the depth and distribution of that value vary considerably (see Table 1). This is consistent with a technology in the early stages of diffusion rather than with speculative mania. In other words, AI does not resemble tulip mania: the underlying technology is creating observable benefits even if only a subset of firms have achieved large-scale P&L impact.

Table 1: Synthesizing Different Research Reports

Step 2: Is the Value Created Commensurate with Market Valuations?

The second question in our thinking architecture shifts from enterprise outcomes to the financial markets. Even if AI is delivering value today, does that value justify its extraordinary market capitalization? Current valuations — across semiconductor firms, cloud providers, model developers, and AI-adjacent industries — implicitly assume substantial future productivity gains.

The simple answer is that we don’t know whether AI is currently priced correctly. Even in a basic discounted cash flow model, small changes in assumptions can create massive differences in valuation.

Consider a simplified example: Suppose an AI company will generate $100 of value in five years, and the risk profile translates to a 10% discount rate. Under these assumptions, the company would be worth about $62 today (see Table 2).

But the truth is, we don’t know when AI will deliver value, how much value it will generate, or how risky the path will be. If we instead assume the company will take twice as long to realize value, produce only half as much, and face double the risk, its present value falls to just $8 — roughly one-eighth as much.

These inputs are guesses, and the range of plausible guesses is huge. That’s why it’s so easy to misprice AI, in either direction. It could be a bubble, it could be fairly priced, or it could even be undervalued.

Table 2: The Sensitivity of AI Valuations to Assumptions

Conclusion: What Kind of Bubble Is AI?

So what kind of bubble — if any — is AI? The evidence suggests that AI is not tulip mania: The value is real, observable, and already present across a wide cross-section of enterprises. Even modest pilot deployments are yielding productivity gains, and early adopters are beginning to integrate AI into meaningful parts of their operations.

While markets are pricing in a future defined by large-scale productivity gains and organizational redesign, firms are still contending with the early, difficult work of integration, which helps explain the limited P&L gains to date. In this environment, the critical question for business leaders is not whether AI will eventually create value — it will — but how long it will take for their industry to realize that value.

We cannot tell you how to allocate your personal investments, nor whether anyone should invest their IRA in Nvidia. But we can offer a simpler recommendation to executives: Invest your time. In many industries, effective generative AI deployment will be crucial to survival — not in the distant future, but in the competitive landscape emerging right now.

Takeaways

If we combine insights from the Wharton and MIT studies, a consistent pattern emerges: Most firms are experimenting with AI, but only a small minority have figured out how to turn experimentation into measurable P&L gains. To join that group, focus on three priorities:

1. Re-imagine: Run a true whiteboard exercise to rethink what your company should become in the AI era — not just how to bolt AI onto what exists, but how AI changes the very shape of your industry.
2. Fuel growth: Don’t use AI simply to make existing experiences cheaper. Use it to make customer experiences better. Efficiency matters, but growth is where the real value lies.
3. Urgency: Act now. The pace of change is already extraordinary. Waiting increases the risk of falling behind, and catching up will only become more difficult.