The AI industry is attracting unprecedented investment, but Nobel economist Paul Krugman warns that most of the money being thrown at artificial intelligence companies may never generate enough revenue to justify the spending. In a conversation with historian Heather Cox Richardson, Krugman explored whether the current AI enthusiasm mirrors past bubbles like the dot-com crash or the 17th-century Dutch tulip mania, and what that means for investors and the broader economy.

What Exactly Defines an Economic Bubble?

A bubble occurs when investors pour money into something with little realistic chance of commercial payoff, sustained primarily by the momentum of others entering the market. Krugman explained that bubbles follow a predictable psychological pattern: people invest because everyone else is investing, prices rise because of increased demand, and the cycle continues until new investors stop arriving. At that point, the structure collapses.

"A bubble is a natural Ponzi scheme. It's something where you get in and you make money because other people get in, and people keep on coming in because everybody before them made money. But in the end, it's a game where the money isn't really there," said Paul Krugman.

Paul Krugman, Nobel Prize-winning Economist

The key distinction is that bubbles aren't simply cases where investors are wrong; they're situations where investors are wrong in predictable ways, sustained by collective momentum rather than underlying fundamentals.

How Does the Current AI Boom Compare to Past Bubbles?

Krugman identified striking parallels between today's AI investment frenzy and historical precedents, though with important differences. The dot-com bubble of the late 1990s offers the closest comparison, as both involve massive capital flowing into technology companies with uncertain revenue models. However, Krugman noted that the AI boom also resembles the California Gold Rush of the 1840s, where the real money wasn't made by prospectors searching for gold but by suppliers selling equipment to them.

In the modern AI context, this analogy holds particular weight. While companies like OpenAI and Anthropic are the high-profile players attracting investor enthusiasm, the actual profits are flowing to infrastructure providers. Nvidia, which manufactures the specialized chips required to train and run AI models, has become enormously profitable by selling the computational equipment that AI companies depend on. Similarly, cloud computing providers are making substantial revenue by renting out processing capacity to AI developers.

Where Is the Money Actually Being Made in AI?

The current state of AI profitability reveals a troubling pattern. Most of the revenue generated so far comes not from AI services themselves but from the infrastructure supporting them. Anthropic, one of the two largest AI companies alongside OpenAI, has begun generating some revenue through subscription services like Claude, its AI assistant. However, this income remains modest compared to the billions being invested in AI development and data center construction.

A critical issue undermining the business model is pricing. Currently, the cost of providing AI services is heavily subsidized. For every dollar of computing power that users consume, companies are subsidizing between $3 and $25 in costs, according to Richardson. This means users are paying only a fraction of what the service actually costs to deliver. The fundamental question becomes whether consumers will ever pay enough to justify the enormous capital expenditure.

Key Factors Distinguishing AI From Previous Bubbles

• Actual Technological Capability: Unlike some past bubbles, AI systems demonstrably work and perform useful tasks, creating genuine utility that didn't exist in purely speculative manias.
• Infrastructure Investment: The AI boom has sparked massive construction of data centers, representing real physical infrastructure rather than purely financial speculation.
• Multiple Competing Players: The market includes OpenAI, Anthropic, Google with Gemini, Elon Musk's Grok, and various Chinese alternatives, creating competition rather than a single dominant player.
• Subsidized User Adoption: Current pricing models require companies to lose money on every transaction, making the path to profitability unclear.

Will AI Companies Ever Become Profitable?

Krugman offered a sobering assessment based on historical precedent. "The most likely outcome is that it will end up being a waste," he stated, noting that history suggests most speculative bubbles do not generate sufficient returns to justify their initial investment. However, he acknowledged that history does not always repeat itself, leaving open the possibility that AI could prove to be a genuine transformative technology with eventual profitability.

Krugman

The critical distinction lies in whether AI companies can transition from subsidized services to profitable ones. This requires either dramatically reducing operational costs or convincing users to pay substantially more for AI services. Neither outcome appears imminent, given current market dynamics and user expectations.

The conversation between Krugman and Richardson highlights a fundamental tension in the AI industry: genuine technological achievement coexisting with questionable financial fundamentals. While AI systems are undeniably impressive and useful, the business models supporting their development remain unproven at the scale of current investment levels.