Five US tech giants' hidden AI debt hits $1.65T, Nikkei says
TL;DR
- The five companies' combined visible and hidden obligations total roughly $3 trillion, nearly double what appears in standard credit models.
- AI server replacement cycles of 18-36 months against bond durations of 5-20 years create a refinancing cliff that standard credit models miss.
- BIS and Moody's each independently flagged the off-balance-sheet AI debt gap before the Nikkei study attached named-company dollar figures to it.
The single number in Nikkei's new study is small enough to read past, and big enough to matter: about $1.65 trillion of off-balance-sheet debt across five US tech giants, tied to the AI buildout.
The context is what makes the number interesting. Nikkei reports that this hidden debt has swelled roughly eightfold in about four years as AI investment ballooned, and it now exceeds the companies' actual reported debt. The specific instruments the study points at are data center leases and GPU supply contracts, obligations that behave like debt if you have to pay them regardless of demand, but that don't sit in the debt line when the market prices the equity.
Meta is the case study the reporting singles out. Its off-balance-sheet liabilities come in around $420 billion, nearly triple its transparent debt. That is a company most investors would still describe as effectively net-cash, when in reality the obligation stack sits roughly on the order of the infrastructure it has committed to build.
The honest caveat is that hidden debt is a construction rather than a GAAP line item, and the retrieved reporting does not spell out per-company figures for the other four giants or the precise methodology, so treat the totals as an analyst estimate rather than settled disclosure. It is also not clear whether the SEC or the ratings agencies will lean on any of this.
What is clear enough to act on: if you were underwriting AI exposure by reading only the balance sheet, you were seeing about half the leverage. The near-term watch is which credit analyst or short seller starts pricing those lease and supply commitments alongside bonds, and whether the disclosure norm follows.
What others are reporting
-
Electronics Weekly Read →
Adds the hardware obsolescence dimension: AI server replacement runs 18-36 months against chip generations, creating an asset-duration mismatch credit models miss entirely.
The server replacement cycle in AI datacentres is now 18-36 months with a new generation of accelerator chips coming out every 18 months.
-
Seoul Economic Daily Read →
Frames the Nikkei findings through a regulatory lens, citing BIS shadow-borrowing warnings and Moody's independently, and names Blue Owl Capital joint ventures as the off-sheet vehicle.
Hidden debt not shown on financial statements is emerging as a new risk factor for the global AI industry.
-
Stratos Research Read →
Deepest structural analysis: explains VIE mechanics, duration mismatch, and names Oracle's $300B single-counterparty exposure to OpenAI as the most concentrated off-sheet risk.
More than $120 billion of data centre spending has moved off balance sheets in under two years
-
Telecompaper Read →
Wire-service relay of the Nikkei dataset; confirms the $1.65T figure and eightfold growth in four years is being picked up across specialist tech trade press globally.
Off-balance-sheet liabilities at five major US technology companies have increased eightfold in around four years to an estimated USD 1.65 trillion
Shared on Bluesky by 1 AI expert
Originally reported by asia.nikkei.com
Read the original article →Original headline: Study: Hidden Debt at Five US Tech Giants Grew ~8x to $1.65T Since 2022, Now Eclipses On-Balance-Sheet AI Borrowing