As tech giants and other corporations invest hundreds of billions of dollars in AI, legal professionals and others caution that a fear of missing out on the next big tech revolution could form a bubble reminiscent of the dot-com era. (iStock.com/Panya7)
Experts contacted by Law360 say the parallels aren't exact, but still a cause for worry.
"It reminds me of the Mark Twain quote, that history doesn't repeat but it rhymes," solo financial derivatives attorney Peter Sanchez Guarda said.
He and others see similarities in the investments early dot-com companies attracted without an obvious path to profit. They also see similarities in the dot-com era's surge in infrastructure buildout, which was funded with circular investments and massive debt.
The late '90s saw an explosion of internet use and startups attempting to capitalize on the world-changing technology, ranging from recognizable e-commerce concepts like Toys.com to an attempt to create an online virtual currency, Flooz.com, and a medical advice website trying to trade on the name of former President Ronald Reagan's Surgeon General C. Everett Koop, DrKoop.com. All three were in Chapter 11 before the end of 2001.
They were not alone. By October 2002 the Nasdaq had declined 78% from its March 2000 peak, erasing an estimated $5 trillion in market valuation, and slightly under half of the dot-coms founded after 1996 were still around by 2004.
At the time, Guarda was with the now-defunct Fleischman & Walsh LLP, a firm that represented Time Warner in its ultimately failed merger with then-internet giant America Online and represented a number of other companies in the telecom and technology sectors.
He said both dot-com and AI investments are influenced by a fear of missing out on new technology and a willingness to embrace "questionable" business models that may not appear to make economic sense.
Then, as now, one of the goals of the smaller companies was to attract a buyout offer by someone bigger, Gavin/Solmonese LLP managing partner Ted Gavin said.
"Every idea is potentially a good idea if you can get to an exit," he said.
Pet supply retailer Pets.com became an emblematic dot-com of the era, thanks to an extensive and expensive advertising blitz featuring its sock-puppet mascot and a strategy of selling basic supplies below cost in an ultimately failed attempt to drive sales of higher-ticket items. The company raised $82.5 million in an initial public offering a month after running a $1 million ad in the 2000 Super Bowl, but by November of that year the company had liquidated.
Guarda noted that losing money with every customer is something Pets.com has in common with current AI providers like OpenAI.
"These AI companies aren't making any money, not even on the highest tier," he said.
But most dot-coms had few material assets and correspondingly low debt loads, which is not necessarily the case for the larger AI players.
OpenAI's business partners — including technology company Oracle and SoftBank — have reportedly taken on $96 billion in debt to fund its operations. Shares of Microsoft, another major OpenAI investor, dropped 12% last month after the company disclosed that about 45% of its undelivered services revenue was tied to OpenAI.
Altogether, AI "hyperscalers" — cloud infrastructure providers including Amazon, Alphabet, Meta, Microsoft and Oracle — issued about $121 billion of debt in 2025, and the sector could see another $900 billion issued in 2026, BNY Mellon said in a December report.
The participants in the internet's early fiber-optic boom also racked up significant debt. Based on estimates that internet traffic was doubling every hundred days, companies like Enron and WorldCom, then the second-largest phone company in the U.S., borrowed heavily in a race to install more fiber-optic cable to meet the expected demand.
"There were times they were digging up the same street six, seven times a year to install more fiber," Guarda said.
However, the demand failed to materialize as quickly as expected. In addition, new technology came along that enabled existing lines to carry more bandwidth, Guarda said.
Enron reported heavy losses from its broadband unit and shut the business down six months before it filed for Chapter 11 in 2001 amid a record-making accounting scandal. The next year, it was revealed that WorldCom had overstated its assets by as much as $11 billion, and that company filed for Chapter 11 in 2003.
Guarda said one dot-com parallel with current AI growth is "circular financing." For instance, fiber-optic makers Lucent Technologies and Nortel Networks — which declared bankruptcy in 2006 — were major financers of the telecoms laying the fiber optics, like the financing that chipmaker Nvidia is contemplating providing to OpenAI, which uses Nvidia's chips.
Guarda said other major AI players like Meta and Google don't need lenders, investors or the retail public, even with spending levels like the $175 billion to $185 billion Google is projecting to spend on AI in 2026.
However, Andrew Odlyzko, a University of Minnesota mathematics professor emeritus and specialist in speculative bubbles, said he has started to become concerned that even these giants may begin to run out of funds and begin to engage in "creative accounting" of the kind that felled Enron and WorldCom. He is particularly concerned that many of the data centers being built to run ever-expanding AI models are being financed through special purpose vehicles — subsidiaries formed to handle specific assets or transactions that keep the debt off the AI developers' books.
"It's a situation where I can't even give you a good explanation of how big or how dangerous it is," he said.
If AI is indeed a bubble, how long it can last is another question. Gavin said he sees "a couple more years of juice" from the trend, while retired U.S. Bankruptcy Judge Judith Fitzgerald of the Western District of Pennsylvania — now a bankruptcy mediator and arbiter — said reports she's read "suggest that the curve may be turning."
One thing that could start a downturn is a shift of public opinion on AI, Fitzgerald said, noting the recent controversy over the use of X's Grok AI to make sexual images of women and children.
"The bigger risk is what AI could do to hurt people," she said.
Guarda said another potential limit is infrastructure — specifically, the supply of high-end computer chips and the ability of the electric grid to feed the ever-hungrier AI models. There is currently up to a four-year wait for new electrical transformers, he said.
There is also the broader effect of higher electric rates driven by AI demand, Gavin said.
"It will absolutely creep over and affect other sectors," he said.
Odlyzko said AI investment seems in some ways detached from reality, saying there's no evidence whether current large language model research will succeed or fail in producing a "general AI" that can deliver on the promises of AI boosters.
"Even the top-notch experts in the area cannot be certain what is happening," he said.
That makes it different from the fiber-optic bubble because back then, people were making "solid, quantifiable arguments" against the claims of doubled internet traffic every 100 days, he explained.
"There was pretty good evidence that wasn't true, but people weren't paying attention to it," he said.
And then there is the question of what will remain if the bubble ends. Pets.com may be long gone, but Amazon started as a dot-com bookseller that didn't turn a full-year profit for nearly a decade. Now, the behemoth is the largest online U.S. retailer. And the "dark fiber" installed during the telecom boom was eventually put to use when web traffic finally rose to meet the early projections.
Gavin — who said he still has a Pets.com puppet in a box in his basement — said many of the dot-coms that had profitable ideas were snapped up by larger companies, and that he expects the same thing to play out among the smaller companies swimming in the AI giants' wake.
"We've seen this repeat itself throughout history," he said.
--Editing by Alanna Weissman.
For a reprint of this article, please contact reprints@law360.com.