Tech's $120B Off-Balance-Sheet Datacenter Debt Isn't Enron 2.0

Tech giants are financing datacenter construction through off-balance-sheet vehicles

Meta, Oracle, xAI, and CoreWeave are collectively raising billions to build AI datacenters through a financing strategy that keeps most debt off their corporate balance sheets. According to reporting from the Financial Times in December 2025, tech companies had shifted more than $120 billion of AI datacenter spending into special-purpose vehicles and similar structures. Goldman Sachs projects hyperscalers could spend $5.3 trillion on AI infrastructure through 2030.

The mechanics work like this: A company forms a separate entity that isn't consolidated in its financial statements. That entity raises capital from investors, banks, and financial firms to construct the datacenter. The parent company then contracts for exclusive use of the facility once built—securing the infrastructure without showing the construction debt as a liability.

Critics warn this resembles Enron's off-balance-sheet accounting that preceded the energy giant's 2001 collapse, which wiped out tens of billions in shareholder value and contributed to a broader market crisis.

Why it matters

The comparison to Enron misses critical differences in transparency, asset quality, and market fundamentals. Today's structures finance physical infrastructure with proven demand—not speculative energy contracts. North American datacenter capacity grew 36% last year while vacancy rates fell to a record 1.4%, according to CBRE's H2 2025 report. Unlike financial engineering that masks operational failure, these vehicles distribute genuine capital requirements across investors who understand the risk profile.

The 1980s biotech parallel offers better context

A more apt comparison comes from biotechnology financing in the 1980s and early 1990s. Companies like Centocor formed limited partnerships to fund drug development, holding minority interests while the partnerships issued debt and took investments from limited partners. Centocor retained exclusive rights to any successful drugs.

Hundreds of millions flowed into these structures. Some drugs failed in clinical trials, but the approach didn't trigger market panic because risks were distributed among sophisticated investors. The financing eventually fell out of favor as cheaper alternatives emerged, but it served its purpose during a capital-intensive growth phase.

Today's risks are different and more recoverable

The current datacenter buildout differs in three fundamental ways from both Enron and the biotech partnerships:

First, disclosure requirements and public scrutiny have intensified dramatically. Companies must provide detailed footnotes about these arrangements, and the investing public has access to far more information than in previous decades.

Second, the underlying assets are tangible and durable. Biotech partnerships funded drug candidates with high clinical failure rates. Datacenters are land, buildings, electrical infrastructure, and computing equipment. Even if a facility underperforms financially, the physical assets retain value and can be repurposed. As Jeff Bezos noted, AI represents an "industrial bubble"—and industrial bubbles leave railways, fiber-optic cable, and infrastructure behind.

Third, genuine market demand exists. Microsoft estimates only 17.8% of the world's working-age population currently uses generative AI, suggesting adoption remains in early stages. Demand for computing capacity continues outpacing supply across major markets.

Some investments will inevitably disappoint. Some lenders will take losses. Some datacenters will be worth less than their construction cost. But that's precisely why these financing structures exist: to distribute enormous capital requirements among investors willing to accept the risk in exchange for potential returns.

The obligations are disclosed, the assets are real, and the market need is demonstrable. That's financial engineering serving its intended purpose—not fraud waiting to detonate.

These details were first reported by The Guardian.