The race to dominate artificial intelligence is creating a financial reality that doesn’t appear on the balance sheets of America’s largest technology companies.
A new analysis by Nikkei estimates that five U.S. tech giants, Alphabet, Microsoft, Amazon, Meta, and Oracle, have accumulated about $1.65 trillion in off-balance-sheet AI-related obligations, a figure that now exceeds their combined reported debt of roughly $1.35 trillion. The commitments are disclosed in financial statement footnotes rather than on the balance sheet itself, making them easy for many investors to overlook.
The findings highlight how the AI infrastructure boom is reshaping corporate finance. Instead of relying solely on traditional borrowing, major cloud providers are locking in long-term data center leases, GPU supply agreements, and server purchase commitments worth hundreds of billions of dollars to secure computing capacity years in advance.
“Hidden debt at U.S. tech giants swelled eightfold in four years to an estimated $1.65 trillion as artificial intelligence investments ballooned, a Nikkei study shows, exceeding actual debt and making it tougher for investors to assess risk,” Nikkei wrote.
For Wall Street, the question is no longer how much these companies are spending on AI. It is how much future financial risk is building behind the scenes.
The Hidden Cost of AI: Big Tech’s Off-Balance-Sheet Debt Tops $1.65 Trillion
The obligations identified in Nikkei’s analysis are not hidden from regulators or auditors. Public companies disclose many of them in the notes accompanying quarterly financial statements, following existing accounting standards. Yet they often remain absent from the debt figures that receive the most attention from investors.
Many of these commitments involve infrastructure that has not yet entered service.
Hidden debt at U.S. tech giants swelled eightfold in four years to an estimated $1.65 trillion (Credit: Nikkei)
Long-term contracts for Nvidia GPUs, future server deliveries, and data center lease agreements that have yet to commence typically stay off the balance sheet until accounting rules require recognition. That treatment can make a company’s reported debt appear far smaller than its total future contractual obligations.
“The five companies’ hidden debt, which does not appear on balance sheets, totaled $1.65 trillion in the most recent quarter, exceeding the roughly $1.35 trillion in debt reflected on their balance sheets. The data includes some estimates.”
Among the companies examined, Meta carried the largest estimated off-balance-sheet commitments at roughly $420 billion, nearly three times its reported debt. Oracle followed with about $273.3 billion, marking an increase of more than thirtyfold over the past four years as the company ramps up AI infrastructure projects tied to OpenAI’s Stargate initiative.
AI infrastructure is becoming one of the biggest capital commitments in tech history
Building AI infrastructure has become one of the most expensive investments the technology industry has ever undertaken.
Modern hyperscale data centers cost billions of dollars to construct. The largest campuses can require tens of billions before a single AI model begins training. To avoid tying up enormous amounts of capital upfront, many technology companies lease facilities from specialized operators instead of owning every building outright.
That approach allows companies to secure computing capacity years ahead of demand without immediately adding equivalent debt to their balance sheets.
Oracle has leaned heavily into that model through Stargate, the massive AI infrastructure project announced alongside OpenAI. Meta has adopted a similar strategy through partnerships with outside investors.
Last year, Meta formed a joint venture with funds managed by Blue Owl Capital to develop a Louisiana data center initially valued at $27 billion. The company later increased its expected investment to more than $50 billion. Under the arrangement, Meta owns a minority stake in the operating company and leases the facility, giving it access to computing resources without directly owning the entire project.
The agreement contains another notable provision. Meta has committed to protecting investors against losses if the lease ends early and the facility becomes unnecessary.
Investors are paying closer attention
Institutional investors have begun scrutinizing these commitments more closely.
Morgan Stanley examined the issue in an investor report, and Moody’s warned earlier this year that lease obligations tied to future data centers were climbing quickly.
The concern is less about accounting treatment than future demand.
The largest cloud providers believe AI adoption will continue climbing for years. Amazon Web Services CEO Matt Garman recently said his company’s AI investments are “not speculative.” Microsoft, Alphabet, and Amazon collectively reported cloud service backlogs approaching $1.45 trillion at the end of the first quarter, giving executives confidence that demand will support continued expansion.
The calculation changes if AI infrastructure outpaces customer demand.
Data centers built for future workloads could operate below capacity. GPU purchase commitments would still need to be honored. Lease payments would continue regardless of utilization levels. Assets that looked indispensable during today’s AI race could lose value if demand fails to match expectations.
A financing model built on long-term confidence
The analysis arrives as spending across the AI industry continues reaching new highs.
Many technology companies are already issuing bonds and raising capital to help finance AI expansion. Institutional investors are supplying additional funding through joint ventures and infrastructure partnerships that keep part of those obligations off the balance sheet until projects become operational.
Economists at the Bank for International Settlements recently described this financing approach as “shadow borrowing,” referring to capital raised through structures that avoid immediate recognition as debt under existing accounting rules. They warned that any slowdown in AI demand could expose vulnerabilities across the data center investment cycle.
The comparison does not suggest accounting misconduct. Companies are following established reporting standards. The broader concern centers on visibility. Investors who focus only on balance sheet debt may miss the scale of future obligations sitting elsewhere in corporate filings.
That distinction matters more than ever as the AI race enters its next phase.
The industry’s largest players are making commitments measured in the hundreds of billions of dollars, betting that demand for AI computing will continue rising for years. If those bets pay off, today’s off-balance-sheet obligations could become tomorrow’s engines of growth. If demand falls short, those same commitments may become one of the defining financial risks of the AI era.
