NEW YORK – SEPTEMBER 17: Traders work on the floor of the New York Stock Exchange (NYSE) September 17, 2008 in New York City. Following news late Tuesday of the U.S. government takeover of insurer American International Group (AIG), U.S. stocks opened lower Wednesday as investors digested the news. (Photo by Spencer Platt/Getty Images)
Getty Images
The generative AI boom grew even bigger this week when Open AI announced its latest multibillion-dollar chip deal, this time with Broadcom – sending its stock up 10%, CNBC reported.
The agreement, along with pacts with AMD and Oracle, is one of many that’s pumping money into an AI bubble that looks like it could end much like the dot-com bust did — sending the Nasdaq down 78%.
If such a value decimation began today, it would ultimately swipe $40 trillion from the Nasdaq’s market capitalization.
Or will the Gen AI boom continue for decades? Read on to learn how to tell which scenario is most likely.
Why Did The Dot-Com Boom Bust? Will The Generative AI Brain Rush Meet The Same Fate?
In March 2000, the dot-com bubble – which began with the 1995 initial public offering of Netscape – began bursting. The collapse subtracted $3.6 trillion from the Nasdaq’s market capitalization.
There was also lots of debt — $500 billion borrowed to build fiber optic networks went largely unrepaid due to the bankruptcy of telecom providers such as WorldCom and Global Crossing.
While the Nasdaq began declining from its peak on March 10, two events on March 20, 2000 lit the fuse to this cash conflagration.
That day “Burning Up,” a Barron’s cover story by Jack Willoughby – who interviewed me for the article – was published. Willoughby’s column listed 51 publicly traded Internet firms he forecast would burn through their cash out of 207 by March 2001.
That same day, MicroStrategy – whose CEO Michael Saylor is now Executive Chairman leading the renamed company’s strategy of selling stock to buy bitcoin – announced a revenue restatement for 1998 to 1999 due to “aggressive accounting practices,” according to the Securities and Exchange Commission.
MicroStrategy’s accounting irregularities further shattered investor confidence. By April, the company also restated 1997 results – news of which sent the stock down 91% from its peak.
The GenAI boom has certain parallels with the dot-com one. There is OpenAI – the cash incinerating startup driving the hype; debt-heavy infrastructure operators such as CoreWeave; and cash rich providers of customer financing – notably Nvidia and Oracle – to encourage those cash-burning startups to buy more of their picks and shovels, about which I wrote in my book Brain Rush.
To build data centers, infrastructure operators are borrowing more than $1 trillion, according to a recent Morgan Stanley study. This could end badly if the cash required to repay that debt is insufficient to cover interest payments.
I see three scenarios: GenAI keeps booming, there’s a soft-landing in which valuations decline somewhat, and the OpenAI bankruptcy scenario which brings it all down abruptly should the company be unable to raise more capital to fund its money-losing operations.
The Undiscussed Debt Bomb
The bursting of the dot-com bubble wiped out more equity – much of it held by individuals because so many dot-com companies had gone public – than debt. But the unpaid debt damaged lenders and other institutions.
For example, WorldCom’s $41 billion bankruptcy and Global Crossing’s $12.4 billion collapse resulted from debt service way in excess of their money losing operations’ ability to repay. During the telecom boom, companies laid 80 million miles of fiber optic cable – between 85% and 95% of which remained unused years after the crash, The Bubble Bubble reported.
Today’s GenAI boom could wipe out a significant amount of privately held equity value. While it does not appear a peak has been reached, the debt taken on to finance the building of AI data centers is expected to be even larger than it was during the dot-com boom.
Major tech companies pledged a record $320 billion in capital expenditures for 2025, Fortune wrote. That number is growing so much that capex is forecast to reach $2 trillion by 2028 – a whopping $1.5 trillion of which will come from various forms of debt, according to Morgan Stanley.
Meta, Amazon, and Microsoft generate massive cash flows. But the growth in debt to build AI data centers highlights the strain on their internal resources. The Morgan Stanley report suggests external financing, including high-cost private credit deals, is necessary to supplement their free cash flow.
Investors should note Oracle’s recent experience: the company is already losing $100 million quarterly on its data center rentals to OpenAI noted AInvest despite signing a $300 billion, five-year deal, as I wrote in a September Forbes post. By the end of fiscal 2028, Oracle is forecast to burn nearly $29 billion in free cash flow, noted a Visible Alpha report featured by the Wall Street Journal.
If Oracle — a relatively disciplined operator— can’t make AI infrastructure profitable, who can?
The Circular Financing Shell Game
The most alarming parallel to 2000 involves circular financing arrangements in which a company finances its customer’s purchases. Nvidia, Microsoft and others are now engaging in this practice.
Here are five examples:
Nvidia invested $100 billion in OpenAI which the ChatGPT provider will use to buy Nvidia chips, CNBC reported. Microsoft, which provides about 20% of Nvidia’s revenue, noted Yahoo! Finance, is OpenAI’s largest investor. CoreWeave, which derives an estimated 60% of its revenue from OpenAI, has Nvidia as an investor and uses GPUs as collateral to buy more GPUs, wrote CNBC. AMD will supply OpenAI with multiple generations of its Instinct GPUs under a 6-gigawatt deal. OpenAI can acquire up to 10% of AMD if certain deployment milestones are met. Word of the deal added some 43% to AMD’s stock market value, according to CNBC.Broadcom will build custom chips and AI computing infrastructure for OpenAI which could add more than $100 billion to Broadcom’s revenue by 2029, noted Bernstein Research analyst Stacy Rasgon. However, if OpenAI stumbles, Broadcom will have made an expensive bet on new technology which may not be “easily transferable to other customers,” noted the Wall Street Journal.
These circular deals are not enough to satisfy OpenAI’s ambition. The company aims to build 250 gigawatts of new computing capacity by 2033 with a price tag exceeding $10 trillion, noted Bloomberg.
What if enterprises – which are consuming much of the data center capacity being built to train and operate AI chatbots – decide they are wasting their money?
With 95% of AI pilot projects failing to yield meaningful results, according to MIT NANDA, that absence of a return on their $40 billion investment could come soon if a recession is near.
March 20, 2000 Redux: The Catalyst Waiting to Happen
What could trigger a bursting of the AI boom? OpenAI needs to reach $125 billion in revenue just to break even—a target not expected until 2029, Reuters reported. Until then, the ChatGPT provider must finance its losses by raising capital.
A loss of confidence in the industry’s ability to borrow money to build more AI data centers could send investors scrambling for the exits. However, given the small number of publicly-traded AI chatbot providers, it is difficult to envision another article like Willoughby’s March 2000 “Burning Up” – which was based on dozens of public companies likely to burn through their cash.
Why This Time Could Be Worse—Or Better
Differences between today and 2000 may make a GenAI crash less severe. Unlike the dot-com era’s 2,288 IPOs, noted International Banker, many of today’s AI companies remain private. More than 300 enterprise software unicorns are trapped without exits. Their median age is “11.5 years, and 50% need to raise or exit within 12 months,” wrote SaaStr.
The relatively small number of GenAI IPOs could limit the damage of a crash. In 2000, retail investors were borrowing money to bet on publicly-traded Internet companies which collapsed after they burned through their cash. Retail investors’ losses spread the dot-com collapse’s pain more widely.
If recently public GenAI companies – such as CoreWeave or SoundHound — were to implode, the relatively small number of GenAI IPOs might cause less damage to retail investors than did the dot-com bust.
Another thing that’s better about today is the high profitability of many of the tech giants betting on AI. Today, the top 10 stocks in the S&P 500 contribute 28.8% of total earnings – a big jump from 2000’s 16.1%, reported Fortune.
Preparing for Multiple Scenarios
Investors should prepare for three scenarios:
The Soft Landing (estimated 35% probability): In this scenario, AI valuations might decline 60% to 70% over a two to three year period without panic. Strong growth and earnings at Microsoft, Google, and Meta might provide cushioning while enterprise adoption of AI would continued at realistic valuations.The OpenAI Bankruptcy Cascade (25%). The most pessimistic scenario starts with OpenAI failing to raise enough capital to cover its $14 billion annual cash burn – leading to emergency measures. Under this scenario CoreWeave could lose 60% of its revenue and file for bankruptcy. Nvidia might suffer a 6% reduction in its revenue and take a $100 billion investment write-down. Oracle and Microsoft would likely mark down their OpenAI investment, noted Where’s Your Ed At. This could lead to an abrupt curtailment of AI capital expenditures – resulting in a recession, a drop of 40% to 50% in Nvidia’s stock price and a 20% to 30% decline in the S&P 500 – since AI stocks provided 75% of recent gains.The Continued Boom (40%). In this optimistic scenario, companies discover ways to use AI to achieve efficiency and growth improvements – making current investments profitable. Revenue growth accelerates to match valuations and the bubble inflates further before any reckoning.
What Investors Should Monitor
To get a sense of which scenario is most likely, investors should monitor these indicators:
OpenAI’s operations. Will OpenAI develop a scalable business model? To know, track statistics such as quarterly burn rates versus revenue growth, enterprise customer retention and churn data. VC funding velocity changes. Will VC bets on AI drop below their record $192.7 billion as of early October, as Morgan Lewis noted? Nvidia customer concentration risk. Will Nvidia become more dependent on a small number of customers? For instance, in August two mystery customers reportedly made up more than 40% of the company’s revenue, noted The Economic Times, and CoreWeave debt coverage ratio. Will CoreWeave struggle to repay its debt? CoreWeave has $11.2 billion in debt at interest rates ranging from 7% to 15%. The company “spent more than $250 million paying interest on that debt last quarter on just $19 million in operating income,” reported Forbes.
When today’s equivalent of Willoughy’s Barron’s article appears — providing specific data on which companies will run out of cash when — the psychological foundation will crack.
The question is not if such a moment comes, but when, and whether the AI boom’s stronger fundamentals can cushion the fall that the dot-com boom’s relative lack of substance made inevitable.
