Steve Eisman, the investor who rose to fame for accurately predicting the subprime mortgage crisis, has recently turned his attention to the booming artificial intelligence sector. In a recent podcast, he revealed that some AI companies are employing a financing method that played a notorious role in the Enron scandal and the 2008 subprime crisis: utilizing special purpose entities for off-balance-sheet financing to fund massive data center construction.
Special purpose entities are typically independent legal entities established by corporations for specific projects, primarily to raise capital without increasing the parent company's on-book liabilities. Eisman admitted he assumed such complex financial maneuvers had vanished with the tightening of regulations, but the reality has left him shocked. He noted that Enron precisely used a massive number of such entities to hide debt and inflate profits, while Wall Street banks prior to the subprime crisis used similar off-balance-sheet tools to accumulate enormous hidden risks. Today, these tactics are resurging in the new AI era on an unprecedented scale.
According to Eisman's observations, the AI boom is fundamentally reshaping the business logic of big tech companies. In the past, firms like Microsoft, Alphabet, Oracle, and Meta enjoyed abundant free cash flows with asset-light models. However, as the demand for computing power surges for large model training and inference, these companies are forced into an asset-heavy race, requiring huge capital expenditures to build data centers and supporting power and network infrastructure. Internal cash flows can no longer meet capital expenditures of this magnitude, forcing them to turn to the bond and credit markets.
This sharp surge in capital expenditure has directly led to rising corporate leverage and pressure on credit ratings. Citing Oracle as an example, Eisman pointed out that it was recently downgraded by S&P to BBB-, just one step away from junk status. Against this backdrop, shifting some debt off-balance-sheet has become a tempting option to maintain ratings. He specifically highlighted Meta's Hyperion data center project in Louisiana. The project issued approximately $27 billion in debt through a special purpose entity set up in collaboration with private credit firms. Although Meta holds only a 20% stake in the entity, it bears the risks of data center construction, a 20-year lease commitment, and project delays and cost overruns. This practice of moving massive debt off its own balance sheet reminded Eisman of the financial engineering that triggered market turmoil in the past.
Eisman emphasized that he is not entirely dismissing the legitimacy of special purpose entities. His core concern is that, with the runaway growth in AI infrastructure investment, the market is becoming increasingly reliant on these complex financing structures to mask true leverage levels. If companies bear the vast majority of the economic risks in substance while achieving debt invisibility on their books, traditional financial metrics will fail, making it difficult for investors to see the true long-term obligations of the companies. Particularly in the current high-interest-rate environment in the U.S., rising financing costs make off-balance-sheet financing more attractive, but also render the potential long-term burdens more dangerous.
Earlier this year, Eisman repeatedly warned the market that the current AI boom is highly dependent on continuous heavy spending by a few tech giants. Now, his focus has shifted from valuation to financing structures. The massive hundreds of billions of dollars spent on data center construction are making AI competition increasingly resemble infrastructure sectors like energy or telecommunications. The key to whether this capital feast can continue may no longer be just algorithmic breakthroughs or chip iterations, but whether these giants can continue to access funds at low costs in the reality of high interest rates. For Eisman, the most unsettling aspect is not the AI technology itself, but the fact that Wall Street's familiar financial engineering tricks are making a comeback dressed in the guise of new technology.
