Is the AI Bubble Worse Than the Dot-Com Bubble? Wall Street Legend Warns: Billions in AI Investments Face Reckoning

Jim Chanos, the legendary short-seller and founder of Chanos & Co., recently stated that the current artificial intelligence (AI) infrastructure investment boom is witnessing capital inflows at a speed and scale unmatched in history. However, the economic logic underpinning this frenzy remains unverified, raising serious concerns.

Chanos pointed out that the market is now playing a classic "story premium" game, where investors are willing to pay a premium for corporate promises while largely ignoring whether the "reality" of financial numbers can deliver.

He warned that hundreds of billions of dollars in capital expenditures are being justified by short-term spot prices, yet are being used to support long-term asset investment decisions spanning two decades—creating a fundamental mismatch in duration.

Chanos compared this AI infrastructure boom to the late 1990s internet infrastructure bubble, calling the current situation "worse."

Citing data, he noted that from 1998 to 2002, the two most capital-intensive industries during the dot-com era—competitive local exchange carriers (CLECs) and fiber-optic deployment—spent a combined total of about $100 billion over five years, averaging roughly $20 billion annually.

In contrast, during this current AI cycle, a single company’s annual spending on data centers and AI infrastructure has already far exceeded the total five-year expenditure of those two industries combined.

Even more concerning is the difference in funding structure. Chanos noted that during the dot-com bubble, the actual corporate customers footing the bills—such as Bank of America (BAC-US), General Electric (GE-US), or Coca-Cola (KO-US)—remained profitable throughout the cycle, at worst only reducing orders.

But in this cycle, only the hyperscale cloud providers can generate profits and self-fund their spending. Nearly all other participants in the ecosystem rely on venture capital or highly leveraged financing to sustain operations.

He likened this investment frenzy to the asset-liability maturity mismatch that preceded the 2008 financial crisis. Back then, financial institutions used short-term repo market funding to support long-term derivative positions, ultimately triggering a systemic collapse.

"This is basically a 'Finance 101' level mistake," Chanos said. "People are using short-term spot prices to justify long-term capital plans."

Accounting Practices Hide True Depreciation Pressure

Chanos specifically highlighted an often-overlooked accounting issue: a large volume of purchased but not yet activated GPUs and data center equipment are mostly recorded on balance sheets under "construction in progress," and have not yet begun depreciation.

"This means these assets are simultaneously facing economic and technological depreciation, yet this is not reflected at all in the income statement," he said.

Analysis suggests that, given an average 18-month gap between GPU procurement and full deployment, even if the book depreciation period is set at 5 to 6 years, the actual amortization period could stretch to 6.5 to 7.5 years.

Chanos revealed that his own firm uses a ten-year depreciation period in modeling, but "even then, we still can't clearly calculate the economic benefits of most data centers."

He further pointed out that this accounting treatment is distorting the macro picture. S&P 500 earnings growth is projected at a staggering 28% this year and about 20% next year—far above the long-term average of around 6% annually. "One reason is that massive capital expenditures on one side are being recognized as revenue on the other, while the spending side opts to capitalize rather than expense them in the current period."

Neocloud's 'Light-Asset Shift' Is Self-Contradictory

This AI boom has also spawned a new wave of 'Neocloud' computing companies operating under a heavy-asset model. However, recent market developments are undermining this narrative.

In an interview, Chanos called out emerging cloud player Nebius (YNDX-US). The company recently announced a shift to a 'light-asset' business model, no longer owning data centers or GPUs, but instead transforming into a 'computing power management service provider'—similar to a hotel franchise headquarters—outsourcing capital expenditures to third parties and charging management fees.

Chanos called this a "significant self-own goal." He said, "For the past two years, these Neocloud players have told everyone that heavy assets are the core competitive advantage, a 'money printer.' Now, one of the largest is saying, 'Forget it, we don't need to own these assets anymore.'"

He further analyzed that both traditional data center operators and emerging Neocloud firms have realized that, due to labor and equipment shortages, new project costs are skyrocketing, and future maintenance capital expenditures will likely far exceed what they previously told investors. "That's why they're now rushing to offload assets as quickly as possible."

He also mentioned that SpaceX's (SPCX-US) xAI data center, under Elon Musk, has recently entered into a series of computing power leasing deals—each worth about $1 billion—with Anthropic and Google (GOOGL-US), both featuring extremely short exit clauses allowing withdrawal in as little as three months. "I think these deals carry a strong promotional flavor."

Interest Rates Are the Real Ticking Bomb

Chanos also pointed out that interest rate risk is the most underestimated systemic threat within the AI infrastructure bubble.

He noted that a large volume of assets—including office buildings, data centers, and warehouse facilities—are currently trading at capitalization rates of 5% to 7%, while the U.S. 10-year Treasury yield is around 4.5%, leaving a very thin spread.

Under these conditions, operators are layering leverage and aggressive mezzanine financing to promise equity investors returns as high as 15%. "Once interest rates rise to 6% or 7%, all of this will collapse, and various assets will start blowing up one after another."

He believes that while the credit market currently shows no clear signs of concern, the situation could rapidly deteriorate once rates clearly approach 5%, with widening credit spreads serving as a key warning signal.

He also observed that the yield spread on CCC-rated junk bonds—the lowest tier—has already begun to widen, though BBB to BB-rated bonds have not yet followed suit.

Additionally, Chanos poured cold water on the widely held "power bottleneck" narrative. He stated that data center power costs account for only about 5% to 6% of revenue—the smallest cost component—and "our country absolutely does not lack electricity."

He expects the asset premium logic built on power scarcity to be fundamentally fragile.

When Returns Fall Below Treasury Yields, CEOs Will Face Reckoning

Chanos' core argument is that the capital efficiency of hyperscale cloud providers is systematically declining, and this will force them to make real strategic shifts within the next 12 to 18 months.

According to his estimates, the incremental return on capital for the five leading firms—Google, Meta (META-US), Amazon (AMZN-US), Microsoft (MSFT-US), and Oracle (ORCL-US)—has dropped from around 40% about a year and a half ago to about 20% today. If capital spending growth continues unabated, this figure could fall further to 10%.

"At that point," Chanos said, "these companies' management teams will face a real question: Should we keep burning cash, or would it just be better to buy Treasury bonds?"

He predicts this reckoning moment will surface as early as late 2026 to 2027, and will be unavoidable.

Chanos cited Oracle as an example, noting that the company has the lowest incremental capital return among peers in AI infrastructure, and its stock has plunged 65% to 70% from its peak.

"Any CEO who is truly serious about their duty," he said, "must be looking at Oracle's fate and telling themselves, 'We absolutely cannot end up like that.'"

Chanos concluded with one sentence summarizing his view on current market pricing logic: "In a bull market, people pay a premium for 'promises'; in a bear market, people only discount 'reality.' We are clearly in the former. Will we move to the latter? I don't know."