The artificial intelligence (A.I.) revolution is accelerating, fueled by an insatiable demand for ever-larger data centers to power complex algorithms and training models. While tech giants like Google and Amazon traditionally funded these massive infrastructure projects with their own vast profits, a new wave of smaller players eager to carve out a piece of the A.I. pie are turning increasingly to debt financing. This shift toward borrowing billions, potentially reaching $1 trillion by 2028, has raised concerns about unsustainable risks within the booming industry.
The rise of debt-funded data centers began with companies like Meta partnering with specialized providers. For example, Meta agreed to purchase $14.3 billion in computing power from CoreWeave, a relatively unknown company that recently went public. CoreWeave’s business model relies heavily on borrowing; for every $5 billion it plans to sell in computing power over the next four years, it needs to take out $2.85 billion in loans.
This trend extends beyond such partnerships. OpenAI, the company behind the viral ChatGPT chatbot, is spearheading a wave of ambitious data center projects. Despite generating billions annually, its CEO, Sam Altman, predicts profitability only by 2029. Yet, OpenAI, alongside partners like Oracle and SoftBank, plans to spend over $400 billion building centers across Texas, New Mexico, Ohio, and Wisconsin — much of this likely financed through debt. The exact scale of borrowing remains unclear, but analysts estimate that Oracle alone will need to borrow $25 billion annually for the next four years to meet its commitments.
Adding further complexity, these projects often involve intricate financing structures. For example, at OpenAI’s first data center in Texas, Oracle handles the computer hardware, while Crusoe builds the physical infrastructure, securing a $15 billion loan from Blue Owl Capital and other investors to cover its portion. Meanwhile, SoftBank and OpenAI are reportedly leaning on debt for facilities in Ohio and Texas.
OpenAI has made other bold moves to secure funds, including selling a massive stake to chipmaker Nvidia for $100 billion and accepting a significant equity package of AMD shares — potentially worth tens of billions more. These deals involve commitments to purchase computer chips from both companies, but these agreements offer an escape clause if OpenAI’s needs change. However, even if such escapes exist, the debt itself could become problematic.
This reliance on debt introduces several risks. First, the collateral backing many loans is often the very computer chips themselves, which depreciate rapidly. Second, a broad range of institutions — from banks and private lenders to companies directly investing in facilities — hold these debts, creating systemic vulnerability across the financial landscape. The opacity surrounding many of these deals makes it difficult to fully assess the potential scale of exposure.
“Leverage in the system is what drives risk,” warns Jeremy Kress, a business law professor at the University of Michigan specializing in financial instability. “And it is hard to know how much leverage is in the system.”
The parallels with the dot-com boom of the late 1990s are striking. The rush to build fiber-optic infrastructure fueled by debt led to widespread bankruptcy when promised returns failed to materialize. While A.I.’s potential is undeniable, experts caution that this current reliance on borrowed money risks repeating past mistakes. The rapid accumulation of A.I.-related debt could have cascading consequences if revenue projections fall short, leaving investors and lenders vulnerable and potentially jeopardizing the long-term sustainability of the industry itself.
