Superpower Daily
University of Konstanz Finds AI Agents Coordinate Up to 1,000,...OVHcloud Cites AI Memory Demand as It Raises New Server Prices...Nvidia Takes a Minority Stake in Cloverleaf to Push AI Data-Cen...Watermarks Remover Tops 14,000 Stars; Claude’s Detector Isn’t P...SemiAnalysis Says AgentX Drove 50-Plus Upstream Fixes for AI Ag...MiniMax H3 Opens Its Video Weights, but Not Where Many Develope...University of Konstanz Finds AI Agents Coordinate Up to 1,000,...OVHcloud Cites AI Memory Demand as It Raises New Server Prices...Nvidia Takes a Minority Stake in Cloverleaf to Push AI Data-Cen...Watermarks Remover Tops 14,000 Stars; Claude’s Detector Isn’t P...SemiAnalysis Says AgentX Drove 50-Plus Upstream Fixes for AI Ag...MiniMax H3 Opens Its Video Weights, but Not Where Many Develope...
Businesspublished

AI Data-Center Debt Moves Off Balance Sheets as Spending Could Reach $5.3T

Special-purpose entities can keep much of a project’s construction debt outside a technology customer’s financial statements, while disclosures, physical assets and demand remain central to the case for the buildout.

By 2 min read
AI Data-Center Debt Moves Off Balance Sheets as Spending Could Reach $5.3T
Sign in to save and react

Listen to this story

The audio brief

About 1:27
0:001:27
Read transcript
More than one hundred twenty billion dollars in AI data-center spending had reportedly been moved off technology companies’ balance sheets by December 2025, according to the Financial Times. The structure is straightforward: a separate project entity raises the money and builds the facility, while a technology customer contracts for exclusive, full use once the data center is finished. Much of the construction debt may stay with that separate entity rather than appearing directly on the customer’s books. Gene Marks cites Meta, Oracle, xAI, and CoreWeave as companies raising billions through arrangements that may keep some long-term obligations off balance sheets. Goldman Sachs estimates that hyperscalers could spend five-point-three trillion dollars on AI and data centers through 2030. That is an estimate, not a committed total, but it shows how much capital the buildout could require—and why private lenders and facility owners are taking on more of the exposure. The risks have not vanished. Projects can underperform, lenders can lose money, and data centers can be worth less than their owners paid. The case for demand is strong: CBRE says North American capacity rose 36 percent while vacancy fell to a record-low 1.4 percent. Microsoft estimates that 17.8 percent of the world’s working-age population uses generative AI. The key question is whether that adoption supports the economics of all these facilities, and whether disclosures accurately capture the obligations being created.

Story brief

3 key points

AI infrastructure financing is increasingly being routed through separate project entities, allowing hyperscalers and model companies to secure exclusive data-center capacity while potentially keeping construction debt off their balance sheets. More than $120 billion had reportedly been shifted this way by December 2025, against Goldman Sachs’ estimated $5.3 trillion AI and data-center investment through 2030. The...

  1. 01

    Goldman Sachs’ $5.3 trillion projection is an estimate, not a committed spending total.

  2. 02

    CBRE reported North American capacity rose 36% while vacancy reached a record-low 1.4%.

  3. 03

    Project entities can fund construction while customers retain exclusive use of completed facilities.

Off-balance-sheet financing can make AI data-center borrowing less visible on a technology customer’s books without making the underlying project disappear. The Financial Times reported in December 2025 that more than $120 billion of AI data-center spending had been shifted off balance sheets; Goldman Sachs estimates hyperscalers could spend $5.3 trillion on AI and data centers through 2030.

Capacity is contracted while financing sits elsewhere

The structure described by Gene Marks puts a separate entity in charge of raising capital and building the data center. The technology company contracts for exclusive, full use after completion, while most construction debt may remain outside that customer’s balance sheet. Marks identifies Meta, Oracle, xAI and CoreWeave among companies raising billions through structures that may keep some long-term obligations off their books.

That approach puts private markets in a larger role in the buildout, according to Goldman Sachs. Its $5.3 trillion figure is an estimate rather than a committed spending total, but it underscores the scale of capital being contemplated through 2030.

Physical infrastructure is not a fraud defense

Marks argues that the Enron comparison goes too far, while still calling for scrutiny of disclosures and consolidation. His case is that current disclosure requirements are significant, scrutiny is intense, and the financed assets are land, buildings, electrical infrastructure and computing equipment. Unlike a biotechnology program that fails clinical testing, a data center can disappoint financially without disappearing.

The risks that remain at the project level

  • Some individual data-center investments may fail.
  • Lenders may lose money on projects they finance.
  • Facilities may be worth less than their owners paid for them.

Tight capacity supports the case, not every investment

CBRE said North American data-center capacity increased 36% in the prior year while vacancy fell to a record 1.4%. It also said demand was outpacing supply in nearly every major market. Those figures establish a tight market, even as they do not assure returns for each facility now being financed.

Microsoft estimates that 17.8% of the world’s working-age population currently uses generative AI. Marks sees that estimate as evidence that adoption may still be early. The live question is whether that demand remains strong enough to support facilities whose owners and lenders bear the risk of weaker project economics.

Sources

  1. theguardian.comAn AI ‘debt bomb’ crisis? No. This isn’t Enron 2.0 | Gene Marks

Related coverage

Continue reading