Big Tech’s hidden AI debt hits $1.65T, Nikkei finds

Off-balance-sheet obligations at the five largest US cloud computing companies have swelled eightfold in four years to an estimated $1.65 trillion, driven by opaque financing arrangements for artificial intelligence infrastructure, according to a new analysis published Monday by the Nikkei Shimbun.chosun+1

The Japanese financial newspaper found that after examining footnotes in recent financial reports of Alphabet ↗ 0.24%, Microsoft ↗ 0.03%, Amazon Amazon.com, Inc. ↘ 0.66%, Meta ↘ 1.80%, and Oracle ↘ 4.21%, their combined "invisible debt" now exceeds their official on-balance-sheet debt of approximately $1.35 trillion.chosun

How Hidden Debt Accumulates

The obligations stem from long-term purchase agreements for GPUs and data center leases that accounting standards allow companies to disclose only in footnotes rather than as balance-sheet liabilities. Under generally accepted accounting principles, companies can defer recognizing these commitments until goods are delivered or facilities are completed.fortune+1

Meta's hidden debt leads the group at approximately $420 billion — 2.8 times its official debt — partly driven by a giant data center project in Louisiana with Blue Owl Capital where total development costs exceed $50 billion. Oracle's off-balance-sheet obligations surged 30-fold in four years to $273.3 billion, fueled by heavy leasing for its Stargate AI data center project with OpenAI.chosun

The findings align with a February report from Moody's Ratings, which flagged $662 billion in future data center lease commitments among the same five hyperscalers that had not yet commenced and thus sat entirely off their balance sheets.fortune

Credit Market Strain

The structural shift in how these companies finance themselves is beginning to register in credit markets. Goldman Sachs's The Goldman Sachs Group, Inc. ↘ 1.26% Brian Garrett noted last week that the firm's hyperscaler bond basket widened by 22 basis points in a single week — a move he described as a warning in the normally sedate investment-grade market. He also flagged a growing divergence between credit default swaps and equity put options on hyperscaler names, suggesting one market is pricing anxiety the other has not acknowledged.finance.biggo

Goldman estimates the hyperscale cloud community will need between $5.5 trillion and $6 trillion in capital between 2025 and 2030. Bloomberg reported on July 10 that the five companies have collectively added $350 billion in on-balance-sheet debt over the past five years, with pledged capital spending reaching $725 billion for 2026 alone.bloomberg+1

Risks Ahead

Industry observers warn that if AI demand underperforms projections, deferred payment obligations could abruptly crystallize into liquidity crises. Unlike commercial real estate, AI data centers rely on semiconductors that depreciate rapidly, meaning weak utilization could trigger massive impairment charges.chosun

The companies themselves remain bullish. As of March, the combined backlog of cloud and other projects for Microsoft, Amazon, and Alphabet reached $1.45 trillion, and Amazon Web Services CEO Matt Garman has stated that "current AI infrastructure investments are not speculative".chosun