返回资讯中心
外部精选
软件工程
#开发者实时流

Are we in an AI Bubble? Lessons from 1873 and 2000

The most common question I’m getting lately is, “Are we in an AI bubble?”

Hacker News 最新greekanalyst2026/10/29 · 19 天前21 分钟阅读

以下正文同步自 Hacker News 最新,版权归原站所有,已转换为易读排版。

Sam Issermoyer, CFA, CFP®

Oct 29, 2026

1

1

Share

The most common question I’m getting lately is, “Are we in an AI bubble?”

Two things can be true at once: AI looks to be incredibly transformative and will change our lives, AND it could be an investment bubble.

Sounds crazy, but past technological revolutions have brought speculative excess, overinvestment, financial collapses, and ultimately tremendous economic progress.

We saw it during the internet buildout in 2000, and we experienced it with railroads in 1873.

Let’s look back, compare it to what’s happening now, and lay out a few possibilities for today.

In the mid-19th century, railroads were changing the economy. Goods could be moved across the country, turning isolated communities into interconnected hubs.

Money poured into building out more railroads. According to historical railroad mileage data maintained by the Federal Reserve Bank of St. Louis, the United States added ~7,400 miles of railroad in 1871.

In an incredibly summarized version (go read the book 1873), railroad investors used large amounts of bonds (debt) to finance the building of new lines.

In 1873, investor appetite for these stopped, and railroad companies ran out of capital to continue building. Jay Cooke, a key financier of railroad bonds, went bankrupt after overextending his own credit to keep supporting the railroad bond market.1

Panic ensued, and by 1875, new construction had fallen to roughly 1,700 miles (down 77%).2

In 1873, the market is estimated to have fallen 23%, with railroad stocks down over 30%.3

Not to be flippant, but a ~20% market drop happens and isn’t the end of the world. Since 2018, we’ve experienced the market falling 20% twice, and coming close two other times:

The market and economy eventually recovered, and by 1882, the United States added more than 11,500 miles of railroad annually. By 1890, the national railroad network had grown to more than 163,000 miles.4

The country continued to benefit from the infrastructure that had been built.

History rhymes, but 2000 was a slightly different tune.

In late 1999, the NASDAQ (an index of primarily tech stocks) went bananas, doubling from 2,400 to over 4,800 by early 2000.

Companies with little revenue and even less profit were attracting billions of dollars. Investors were buying into the idea that the internet would transform the global economy. Companies were valued based on clicks, not revenue or potential profits.

As we all experienced, the NASDAQ then promptly fell over 77%5:

Source: https://datatrekresearch.com/nasdaq-2000-bubble-comp/

The S&P 500, which had fewer tech stocks, was still impacted, falling almost 40%.6

The economic impact wasn’t long-lasting: GDP (a measure of the economy) fell ~0.3%, and unemployment hit 6.3%.7 Not fun, but not the end of the world. You could argue that stocks were too highly elevated, didn’t deserve to rise as much as they did, and that the market recovered without too much damage.

Amazon survived (barely). Google grew into one of the world's most valuable companies. The excess fiber-optic infrastructure built during the boom eventually supported decades of growing internet traffic.

A lot of investors lost money. But the economy ultimately benefited from the infrastructure they financed.

Do you see a pattern from two previous technological economic paradigms?

The infrastructure spending on AI is roughly three times larger than the Dotcom buildout (aka telecom and fiber) and double the 19th-century railroads:

Source: WSJ

That’s a slightly terrifying thought, right?

Key differences separate the past from now. The railroads were a new industry with no cash flow; it was all about future growth. The same thing happened with the dotcom bubble, except internet companies' valuations were absolutely bananas.

Now, history does rhyme, but a few key differences remain.8

The AI companies (except for Anthropic and OpenAI; we will get to those in a minute) have larger operating cash flows:

Source: WSJ and CWA

Operating cash flow is the cash a company generates from its regular business activities before accounting for future investments.

$660 billion in cash flow is massive and very different from the dot-com bubble, when revenue was scant, and valuations were sometimes based on website clicks.

What has the market worried about is that all of the cash flow has gone into investing in the future:

Source: @puckrin

From their perspective, these companies are making the rational decision that the AI buildout is paramount to future profits.

Currently, the market’s growth has tracked along with earnings, unlike in 2000:

Source: @SoberLook

Concerns: Who is Paying?

My main concern is who will eventually pay for AI and generate the profits needed to support this level of AI spending.

Derek Thompson recently wrote a fascinating article titled The AI Story Wall Street Doesn’t Want to Hear (check it out here).9

From his article and guest podcast, more than 55% of businesses on Ramp’s platform paid for at least one AI model in September.

First, AI spending is incredibly concentrated. The top 1% of spenders account for roughly 80% of spending in Ramp’s data. Many of these companies are venture-backed technology businesses.

Second, spending growth may be slowing. Ramp’s research suggests spending among the largest customers has begun to level off, even as usage continues to rise. For example, Uber blew through its entire 2026 AI budget in four months and capped spending.10

Third, AI is getting cheaper. The effective price businesses paid per million tokens reportedly declined 41% between March and August.

Are we in a bubble? Maybe?!? Not satisfying, I know.

I don’t think we are in a mania bubble, but I’m not sure the enormous AI investments will pay off (I reserve the right to change my mind as information on the AI companies comes out).

Railroads transformed the country, and the internet transformed the world.

In both cases, investors correctly predicted the technological revolution but often misjudged the financial returns.

We might build too many data centers. Some AI companies might struggle to turn a profit. Investors could lose substantial amounts of money even as the technology becomes more useful.

Predicting when a bubble will burst is extraordinarily difficult. An investor could trade in and out of their portfolio multiple times a month, driven by fear, enthusiasm, and headlines.

First, a great company isn’t always a great investment. The price you pay matters, even for businesses that fundamentally change the world.

Second, diversification still matters. If the AI trade reverses, great companies can still do well. For example, look at Walmart. Sure, not a sexy stock pick. Three years ago, the stock was at $50, and now it’s at $120.11

Third, watch the cash flow. Long-term winners will be the companies that can turn AI adoption into sustainable earnings and returns on invested capital. Which ones will that be!?

Kevin Kelly has a unique outlook on life and AI. Check out one of his fantastic articles on the most improbable life:

I also thoroughly enjoyed Raymond Carver’s Cathedral, a collection of ~12 short stories about ordinary people struggling with loneliness, strained relationships, and the limitations of communication.

There’s often a gap between what the characters can see and feel and, more importantly, express. Odd occurrences appear in the stories and complement their themes in unexpected ways. I’m still thinking about several of them.

Federal Reserve Bank of St. Louis — Historical Railroad Mileage

Federal Reserve Bank of New York — The Long Depression and Panic of 1873

Wall Street Journal — The AI Build-Out Is Becoming the Biggest Economic Bet in U.S. History

Reuters — AI Investment Boom Puts Big Tech’s Free Cash Flow Under Pressure

Derek Thompson — The AI Story Wall Street Doesn’t Want to Hear

A recession did happen as banks and businesses went under.

Sources: A New Historical Database for the NYSE, 1815–1925 and Unequal Legacies of Financial Crises: Evidence from the Panic of 1873

Federal Reserve Bank of St. Louis — Historical Railroad Mileage

Source: BLS — Unemployment Following the 2001 Recession

9Facts below are from his article: Source.

正文由 探How 从来源站点 RSS 同步,内容未经改写;遇到排版缺失或需要图片、视频时请以原文为准。

打开原文
去社区讨论 →