David Godes remembers his first year as a Harvard Business School professor, when young graduate students started dropping out like flies. It was 2000, the dawn of the modern internet, and would-be Harvard MBA grads thought they’d be better off starting and joining nascent dot-com companies.
They didn’t know it was all a bubble of historic proportions.
“It was crazy,” recalled Godes, whose class of more than 100 quickly shrunk to about 80 that year. Even faculty left academia to get in on the early internet frenzy. It was “a FOMO thing,” he said. “You know, ‘I’ve got to be part of this. All my friends from undergrad are part of startups.’”
Investors ultimately threw too much money at risky startups like Pets.com — pushing their stocks far above levels justified by their underlying businesses. Eventually it all came crashing down, with the bubble burst leading to trillions in lost market cap before the early-2000s recession.
Today’s craze over generative artificial intelligence is different, Godes said. He now teaches at Johns Hopkins, and his students aren’t leaving for Silicon Valley any time soon. They’ve got a healthy skepticism of the emerging technology, he said. That’s just one reason why he sees excitement about AI as entirely unlike the early internet era.
Generative AI has enthralled investors to the tune of many billions of dollars over the last year. Companies that make AI hardware and software, especially the chip giant Nvidia, have seen their stocks skyrocket. It’s led skeptics to warn of another tech bubble that will inevitably burst. One economist said earlier this year that the AI craze has companies even “more overvalued” than in the late-90s.
But to those on the side of the debate, the sense of alarm is short-sighted.
“When we had the internet bubble the first time around… that was hype. This is not hype,” JPMorgan Chase CEO Jamie Dimon told CNBC in February. “It’s real.”
AI hype vs. dot-com hype
“Generative AI is the most disruptive technology since the internet,” said Gil Luria, an analyst with D.A. Davidson.
But there’s a skepticism about AI that’s unlike the dot-com era, Godes said. Much of the political and cultural conversation about AI is doom and gloom: State-sponsored groups using it to meddle in elections, chatbots sending disturbing messages, AI making music that imitates real artists — and of course, the ongoing debate over whether AI will take people’s jobs. (It’s complicated.)
“It’s sort of more of a sense of dread than a sense of wonder,” Godes said.
People were skeptical about the internet, too. But now, with the evolution of the internet as a cultural frame of reference, fears about AI’s downsides are more defined. Governments across the globe, academic institutions, and even companies making AI software are studying its potential risks with a level of scrutiny that wasn’t present in the 1990s.
The dot-com hype was so big that by the spring of 1999, one in 12 Americans surveyed said they were in the process of starting a business. The bubble started to form in the mid-1990s and burst in 2000. There was a massive influx of cash for internet-related tech companies as global interest in personal computers and the World Wide Web exploded. It all happened as the U.S. was experiencing its longest period of economic expansion since the post-World War II era.
Read more: What “bubbles” are and why they happen
Internet companies including Priceline, Pets.com, and eToys went public, captivating investors who sent their market values to soaring heights — all while ignoring their shaky business fundamentals. Banks had a lot of cash as the Fed kept printing money in 1999, and they shoved that money into those same dot-coms. That fall, the market caps of 199 internet stocks tracked by Morgan Stanley were valued at a collective $450 billion — even as their actual businesses lost a combined $6.2 billion. Pets.com went bankrupt less than a year after it went public.
“There were websites in the late 90s that just made no sense,” Godes said. “There was nothing complicated about the technology.”
AI startups are different, he said, because the technology is quite complicated.
“It’s harder for an MBA student without technical training to put together a business plan and go out there and start [an AI] business,” he said.
An ‘AI bubble’ or an ‘AI hardware bubble’?
D.A. Davidson’s Gil Luria takes issue with even using the word ”bubble” for AI. Assets can become inflated and enter a bubble, he said, while their underlying technology goes through cycles. Like all new technology, AI may be in a “hype” phase. But that doesn’t mean all AI-related companies’ values are over-inflated, Luria said.
There’s an important difference between stock rallies for AI hardware companies and those of AI software producers, Luria said. While the share prices for a handful of companies, especially Microsoft, have gotten big boosts from AI, that’s because AI software actually boosted their profits — unlike the websites of the dot-com boom. Today’s AI software stocks are still “trading reasonably within range of their historical [price] multiples,” Luria said. (In other words, while Big Tech’s stock prices are a lot higher than they used to be, their price to earnings, sales, and free-cash-flow ratios aren’t radically different.) And the software those companies make will continue to boost sales for years.
But “hardware is a one-time sale,” he said, so “the bigger disappointment could be in the hardware stocks.” Luria likened the AI chipmaking giant Nvidia to Cisco Systems, a company whose products helped build the early infrastructure of the internet — and whose burst came to define the dot-com era. Nvidia’s chips are to AI what Cisco’s networking hardware was to the early internet, Luria said.
“We had enough tools by 1999 and 2000. We had enough equipment and fiber and routers to support the growth of the internet for years to come,” Luria said. “And that’s what we believe is the point in time we’re at now. … By the end of this year, Microsoft, Amazon, Google, and the like will have enough [AI chips].”
Read more: Google and Intel are challenging Nvidia’s AI chip dominance. It won’t be easy
Cisco stock plummeted 80% between 2001 and 2002 when revenues fell short of expectations, as demand for its networking hardware sunk from record heights. Like with Cisco, Luria said, demand for AI hardware won’t continue at its breakneck pace.
“If investors are counting on the current growth rates for equipment hardware that supports the growth of AI to continue,” he said, “they may be disappointed.”
He pointed to Nvidia’s own biggest customers making their own AI chips. Just this month, Google and Meta, two of Nvidia’s top five buyers, released the latest iterations of their own custom AI chips. While Meta’s isn’t powering its AI applications just yet, Google’s AI chatbot Gemini is being run on its new chip. Because Nvidia’s top five customers make up two-thirds of its revenues, Big Tech shifting its AI hardware in-house could seriously hurt Nvidia’s bottom line, Luria said.
Even bubble believers think AI is no dot-com
Even among the experts who see an AI bubble forming, many say it won’t end as bad as the dot-com burst. Richard Windsor, founder of the research firm Radio Free Mobile, said people are using “convoluted and untested methods to justify very high valuations for [AI] companies,” like they did during the dot-com era.
But, he said, “the internet bubble bursting [was] worse than the AI bubble bursting” will be. That’s partly because even in its “immature form” today, AI is capable of generating substantially greater revenues than the internet was in the 1990s and early 2000s. The internet in the 1990s was “super slow,” he said, “and it took a long time to realize” its full potential. Meanwhile, Windsor said he sees AI’s full potential as ultimately limited. Even if the AI bubble bursts, “what the internet became will be bigger than what AI will become in its current form,” Windsor said.
Windsor said one of the reasons he sees AI models as ultimately limited is that machines can’t tell the difference between causality and correlation.
“Because of that, they will never really get to the point where they can be super intelligent, because they cannot reason,” Windsor said.
Read more: Is Nvidia stock in a bubble that will burst? Wall Street can’t make up its mind
Windsor said he doesn’t know when the AI bubble will burst, but there are signs to look for, including price erosion — or when the price of a product falls over time due to customer demand and competition. Windsor said he is already seeing indications of price erosion starting to take hold. Those signs include OpenAI letting people use its products without an account, which Windsor said looks like the company trying to get more users, and the search engine Perplexity AI starting to sell advertisements despite previously saying search should be “free from the influence of advertising-driven models” — which Windsor sees as a sign its monetization hasn’t gone well. He also pointed to surveys indicating large companies are wary about the deployment of generative AI, due largely to safety and security fears.
“The general expectation out there in the market at the moment is artificial general intelligence is on the way,” Windsor said. “I respectfully disagree with that statement.”
Luria sees Nvidia stock coming back to Earth in 12 to 18 months.
“We may not see the top of the hype until maybe even next year,” he said. “But when we do, there’s going to be a lot of people that are going to be very disappointed.”