Enron collapsed because of outright fraud. Off-balance-sheet financing itself has existed for decades without causing anything close to that.
That's the core of accountant and columnist Gene Marks's pushback against growing "AI debt bomb" warnings, the concern that Meta, Oracle, xAI and CoreWeave are hiding the true scale of their data center debt through unconsolidated financing vehicles.
How the Financing Actually Works
The mechanism is straightforward. A company forms a separate entity to build a data center, that entity raises money from outside investors and lenders, and a contract gives the parent company exclusive use of the facility once it's built.
Because the entity isn't consolidated into the parent's financial statements, most of the debt never appears as a liability on the parent's own books.
The scale worrying critics is real. The Financial Times reported in December that tech companies had shifted more than $120 billion in AI data center spending off their balance sheets this way, and Goldman Sachs estimates hyperscalers could spend $5.3 trillion on AI infrastructure through 2030.
Marks Has Seen This Movie Before
Marks says he watched a similar structure play out decades earlier, without disaster. In the 1980s, as an accountant for biotech firm Centocor, he saw the company and peers like Genentech and Amgen fund drug development through limited partnerships that kept debt off their balance sheets too.
Some of those drugs failed in clinical trials, a far higher failure rate than physical infrastructure typically carries. No market panic followed.
"It's highly unlikely that fraud at that level is being perpetuated now by these companies," Marks said, drawing the direct contrast with Enron that critics have reached for.
The Case That This Time Really Is Different
Marks argues today's version actually carries less risk than the 1980s biotech comparison, not more. Disclosure requirements are far more rigorous now, scrutiny is constant, and investors are better informed than they were before the internet and 24/7 financial media existed.
The underlying assets matter too. A failed drug trial produces nothing, while a data center that disappoints financially is still a physical building full of real equipment.
That distinction is part of why Jeff Bezos has described the current buildout as an "industrial bubble" rather than a purely financial one, the kind that leaves behind usable infrastructure the way railroads and fiber-optic networks did after their own overbuilt eras.
Marks also points to demand data that cuts against a glut narrative. North American data center capacity grew 36% last year, according to CBRE's North America Data Center Trends report, yet vacancy fell to a record 1.4% over the same period.
That's a strange kind of oversupply. A genuine glut would show rising vacancy alongside rising capacity, not a record low. Marks also cites a Microsoft estimate that only about 18% of the world's working-age population currently uses generative AI at all, framing current demand as early-stage rather than saturated.
"I see financial engineering, yes. I don't see a debt bomb," Marks concluded.
What This Means for Miami
This is a genuine, informed counterargument to the concentration risk framing MAIN has covered extensively this year, including J.P. Morgan's Bill Eigen's warnings and Apollo economist Torsten Slok's research on how AI's profitable upstream layers depend on continued financing at the unprofitable model layer.
Neither view has settled the question. Marks and Eigen aren't actually disagreeing about the facts, off-balance-sheet financing is real, the sums are enormous, and some investments will lose money.
They disagree about how much that structure should worry investors given how different today's disclosure environment and asset types are from prior speculative bubbles. For Miami's investors weighing AI infrastructure exposure, that's a genuine, unresolved debate among credentialed observers, not a question with an obvious right answer yet.