He Ate a Year's Salary.

Added: 2026-05-16

[00:00:00]It's 1637.

[00:00:03]You are a sailor in the Netherlands.

[00:00:06]You sit down with a merchant after a long journey.

[00:00:09]He offers you a meal. There is bread, there is fish, and on the table, something that looks like an onion. You pick it up and eat it.

[00:00:18]The merchant goes pale.

[00:00:20]What you just ate is a Semper Augustus tulip bulb. At the time, one of the most valuable commodities [music] in Europe, worth roughly what someone would earn in an entire year. The merchant is furious.

[00:00:32]You just consumed [music] a small fortune. The bulb reflects something extraordinary taking place in the Dutch economy.

[00:00:39]Tulip prices had risen dramatically.

[00:00:42]Everyone believed someone else would pay even more for these onion-looking fortunes. And then, it collapsed.

[00:00:49]Today, we are spending billions of dollars on artificial intelligence, [music] and a growing number of economists are asking the same question the merchant should have asked in 1637. [music] Is anyone actually going to pay more?

[00:01:01]So, the question is, is AI the modern tulip? Is AI a bubble?

[00:01:10]>> [music] >> In 2025, global venture capital in AI reached $258 billion.

[00:01:20]More than 60% of all venture capital on Earth went to a single technological category.

[00:01:26]Amazon, Alphabet, and Meta are committing tens of billions to infrastructure justified by future demand.

[00:01:33]American investor and financial writer Paul Kedrosky has studied every major capital boom in modern economic history, and he has a warning.

[00:01:42]Railroads, fiber optic networks, real estate. In each case, investment raced ahead of demand. The result was overcapacity, financial stress, and in many cases, bankruptcies. To understand whether AI is a bubble or not, we must first understand what exactly is a bubble.

[00:02:00]A bubble exists when the price of an asset rises above what its fundamentals can justify.

[00:02:05]But fundamentals are not always obvious.

[00:02:08]Imagine you were considering buying a vending machine. You expect it to generate $100 in the first year, 95 in the second, and 90 in the third. Once those 3 years pass, the machine becomes useless.

[00:02:20]How much would you pay for it?

[00:02:22]Roughly the sum of those expected earnings, $285.

[00:02:25]That is what economists would call the present value of the machine's future cash flows, its fundamental value.

[00:02:32]Now, suppose someone offers it to you for $600. You would probably say no.

[00:02:37]Then you hear your neighbor bought one for $600 and sold [music] it for 750.

[00:02:42]Suddenly, $600 sounds reasonable. And then someone pays $900 and sells it for 1200.

[00:02:48]Until one day nobody wants to buy it, and the last person holding the machine paid $1200 for something worth $285.

[00:02:56]That is a bubble.

[00:02:58]A chain of individually rational decisions that collectively produce an irrational outcome.

[00:03:04]This immediately raises a deeper question.

[00:03:06]If markets are filled with intelligent investors, why would such deviations persist? American economist and Nobel laureate Eugene Fama has a simple answer. Bubbles should be rare.

[00:03:18]In his 1970 paper on efficient capital markets, Fama argues that prices adjust quickly to available information.

[00:03:25]If something is overpriced, someone will sell it.

[00:03:28]Prices adjust rapidly toward their fundamental values. Fama is so skeptical of bubbles that he argues that we only call something a bubble after it crashes, which means the word explains nothing.

[00:03:39]Another American economist and Nobel laureate Robert Shiller disagrees.

[00:03:44]Shiller developed the idea of irrational exuberance into a theory of speculative bubbles driven by narratives, by the fear of missing out, by stories people tell each other about the future.

[00:03:55]He warned, just before the peak, about the unsustainability of the dot-com boom.

[00:04:00]There is a third perspective from the Austrian School of Economics. When interest rates are kept artificially low, money becomes cheap. Cheap money encourages bets that only make sense under those conditions.

[00:04:12]When bets rise and reality sets in, many of those bets collapse.

[00:04:17]Brazilian American economist Jose Scheinkman of Columbia University offers a more technical approach.

[00:04:23]>> [music] >> Scheinkman argues that bubbles can emerge from asymmetry. It is easier to bet that prices go up than bet that they fall.

[00:04:31]Pessimists cannot act on their beliefs as effectively as optimists, simply because of structure.

[00:04:37]The result is an upward bias in prices.

[00:04:39]So, what do all these perspectives have in common? They basically argue that bubbles can emerge from uncertainty, incentives, and the architecture of markets themselves, which makes what is happening in AI considerably harder to dismiss. So, let's take a look at AI.

[00:04:56]Largest technology firms are committing tens of billions to infrastructure, data centers, chips, and computational capacity.

[00:05:04]Kadrosky calls this the great capex boom pattern. He has seen versions of this before.

[00:05:09]>> [music] >> In the 1840s, British investors poured capital into railroads faster than the economy could absorb.

[00:05:16]In the late 1990s, telecom companies buried fiber optic cable across continents. Enough cable to meet demand that didn't materialize until a decade later.

[00:05:26]In both cases, the technology and the demand were real, but the timing was wrong.

[00:05:31]And the investors who bet on the present pay for a future that arrives late.

[00:05:35]So, are the gains from AI models slowing down? Are the costs of training them still rising? If both are still true, the economics of the current boom become very difficult to justify.

[00:05:47]But there's a problem.

[00:05:48]Tulips didn't do anything. They were just beautiful.

[00:05:52]Whereas, artificial intelligence is generating real value.

[00:05:56]Firms are adopting AI tools. Workers are incorporating them into their daily tasks. Surveys indicate meaningful levels of adoption, even if the diffusion is not yet universal.

[00:06:06]Erik Brynjolfsson, Danielle Li, and Lindsey Raymond at Stanford University studied customer support agents who used AI tools and found that their productivity increased by 15% on average, with even larger gains for less experienced workers. There is also evidence on the revenue side. Nvidia, the semiconductor company at the center of the AI hardware boom, reported that data center revenues exceeded $100 billion.

[00:06:33]These are realized sales, not speculative projections. They indicate that demand for AI infrastructure is already substantial. At the macro level, the picture is more [music] complex.

[00:06:44]Daron Acemoglu, a Nobel Prize-winning economist at MIT, estimates that the aggregate productivity gains from AI over the next decade may be relatively modest compared to some of the most ambitious claims. That might sound pessimistic, but it also suggests something important. Modest is not zero.

[00:07:02]The other important question is whether today's prices already assume it has created all of its possible value. This is what makes AI fundamentally different from tulips and fundamentally similar at the same time. AI is a general-purpose technology. Like electricity or the internet, its full value depends on complementary investments. Firms need to reorganize workflows. Workers need to adapt. Infrastructure must be built.

[00:07:26]>> [music] >> These processes take time.

[00:07:28]The benefits of the technology are not instantaneous. [music] So, where does that leave us? Peter Garber, an American economist known for studying historical financial episodes, offered an important warning. He argued that many famous bubbles had plausible justifications with high prices for reasons that made sense.

[00:07:46]>> [music] >> What changed was not the initial optimism but rather timing. This perspective is particularly useful when thinking about AI. It allows [music] us to hold two ideas at once. The technology can be genuinely transformative and financial markets can still misprice [music] it. These are not mutually exclusive possibilities. Now, back to the initial question. Is AI a bubble that will collapse the world? The most honest answer is that we do not know. But we can ask better questions.

[00:08:14]Are current valuations grounded in realistic expectations about future profits? Are firms investing in the right parts of the technology stack? Is infrastructure being built at a pace that matches adoption? Or are expectations running ahead of what the economy can absorb? History suggests that technological revolutions rarely unfold in a clean, linear way.

[00:08:37]They come with excitement, with overconfidence, with mistakes. But they also come with progress. Tulips did not transform the global economy. Artificial intelligence [music] just might.

[00:08:50]If this story of how ideas, institutions, and innovations sparked your curiosity, >> [music] >> explore more at fee.org and dive into our archive of articles, >> [music] >> browse our free ebooks, or subscribe to our newsletters.

[00:09:09]>> [music]