The disturbing comparisons between the AI ​​madness and the dot-com bubble

stock Cisco Systems (Cisco Systems), the champion of the dot-com era, which at the peak of March 2000 became the company with the highest value in the world, this week returned to trading at the same level – for the first time since that period. This is a cautionary tale of how far stock prices can stray from reality.

The bullish analysts spend a lot of time denying the “90s-style bubble” – which is re-inflating around the artificial intelligence industry. But it’s worth going through and reviewing what’s similar and what’s different between then and now.

How much are the shares worth?

There are many ways to value stocks, and almost all of them present US stocks as the most expensive since the dot-com bubble. The forward earnings multiple (forward P/E), the price-to-cash flow multiple, the “Fed Model” calculation of the excess reward offered by stocks versus bonds, and the CAPE multiple (the cyclically adjusted multiple) – all “scream” that the stocks are expensive.

What do all these have in common? Investor behavior – just like in 1999-2000, they are betting that new technology will provide much faster growth than usual in profits. If it does happen, it does justify a higher value.

Just as dot-com companies were priced based on the hope that the Internet would bring with it a new era of profits from unproven business models, so too with AI. Generative AI has brought chatbots and creating images that look almost like magic – but currently AI is priced well below its production cost – leading to big losses. There is only one difference: many of the dot-com companies of the time showed no revenue, while the AI ​​companies at least sell something.

A heavy financial investment

The Internet was built on a global network of fiber optic cables laid by telecom companies, which led to heavy corporate spending financed by debt. The large language models behind advanced AI are built on huge server farms (data centers), which leads to heavy corporate investments – which are financed more and more through debt, as well as from the cash of the big technology companies from their traditional businesses.

The numbers in 2000 were enormous: over 100 billion dollars were “burnt” on new communication networks in the late 90s. A large amount of optical fibers were laid, so much so that a large part of them were shut down for a decade – until Internet traffic increased to the extent that justified their activation.

The race to build data centers has become even more extreme, with talk of trillion-dollar investments by leading AI developers. The spending is so great that economists say it accounts for a significant share of GDP growth.

In general, companies that provide the tools for those chasing the gold rush in both eras have been exceptionally successful. In 2000 it was Cisco, the manufacturer of the routers needed to connect the Internet; And the telecom companies, with the sector raising profits by about a quarter in the last year of the bubble. Today it is Nvidia and other chip manufacturers that provide the processing power to the data centers and reap the profits.

Nvidia’s growth is better than Cisco’s, but both are exceptional: in 1999 Cisco’s revenue grew by more than 40%, and in 2000 by more than 50%. Nvidia, in the two periods of the last 12 months, managed to reach a growth of more than 150%, and then, in the 12 months to October, it reached a growth of 60%. Such growth naturally excites investors.

Market with a magic word

In 1999, more stocks in the S&P 500 fell than rose. So far this year, 183 stocks, or 37% – are down. However, everything related to AI such as: chip manufacturers, power manufacturers, manufacturers of equipment for building data centers – goes up, while a large part of the rest of the market goes down. In 1999, if you were an “Internet stock” – you flourished. And if not, no one was interested. Much the same is happening today with AI.

Here, too, individuals dominate stock trading and once again bet big on small companies and lose. Both in the bubble of 2000 and in the bubble of 2021 (SPACs, cleantech, crypto and cannabis) – small losing stocks were overtaken in value by profitable small stocks, which less excited the investors.

This is reflected in the Russell 2000 index of small companies, which exceeded the S&P 600 index (which requires companies to show profitability to be included), by 10 percentage points in the 12 months to mid-October of this year, before the trend reversed. The Russell small-cap index exceeded the S&P 600 by such a wide margin only three times before: 1999-2000, 2020-2021, and the rally from the post-dot-com bubble slump in 2002.

Such a trading boom boosted Robinhood Markets by 220% this year, just as the dot-com trading boom boosted E-Trade Financial (now part of Morgan Stanley) by 261% in 1999.

Everyone is talking about a bubble

There are those who say that if everyone is talking about a bubble, it is a sign that there is no bubble. Still, debates about “is this a bubble” raged already in 1999, even while it continued to inflate. The same is happening now. But there is one big difference between the current period and 1999-2000 and that is the intensity of the increases. Yes, Nvidia cost a lot (really, a lot). But a 54% increase this year through the October peak, and 30% in 2025 including the decline since then, pales in comparison to the spikes of 1999. By contrast, Cisco more than doubled in four months from mid-October 1999, while Apple rose 150% that year and Intel jumped 75% in less than three months in early 2000.

Even the best big stocks this year like Western Digital, Seagate and Micron Technology “only” tripled or quadrupled to their highs this week, before falling hard. In 1999, Qualcomm jumped 2,620% – only in 2020 did it return to the level from which it crashed later.

Ultimately, the question of whether this is a bubble depends on whether it will burst. If it turns out that artificial intelligence cannot deliver both the productivity promises and fat profits for its creators, the parallel will be to the painful period after the dot-com.