$1.65 Trillion Off-Balance-Sheet Bomb! The “Invisible Debt” of Five Tech Giants Soared Eightfold in Four Years — Could the AI Arms Race Trigger the Next Liquidity Crisis?

According to Zhitong Finance APP, the frenzied “arms race” among America’s five major tech giants in artificial intelligence (AI) infrastructure is giving rise to a “hidden debt empire” far exceeding their reported liabilities.

Recently, an analysis of the latest financial statement footnotes of Alphabet (GOOGL.US), Microsoft (MSFT.US), Amazon (AMZN.US), Meta (META.US), and Oracle (ORCL.US) revealed that the combined “off-balance-sheet debt” accumulated through long-term purchase commitments and data center leases has ballooned to $1.65 trillion. This not only surpasses their officially disclosed liabilities of around $1.35 trillion but has also soared eightfold over the past four years.

This massive scale of off-balance-sheet obligations is raising increasing concerns on Wall Street and among global regulators: should AI demand fall short of expectations, these future payment commitments—hidden in the financial statement footnotes—could quickly morph into “debt bombs” that devour cash flow in a very short time.

Deferred GPU Purchases Bury Hidden Debt Risks

“Hidden debt” refers to future payment obligations arising from investments, equipment purchases, or long-term leasing agreements, which, under current accounting standards, can be deferred from recognition as liabilities until the actual assets are received or facilities completed. Such obligations neither immediately consume cash nor appear in the main columns of the balance sheet, but are instead disclosed only in small print as notes at the end of financial reports.

Currently, as the global race for AI dominance intensifies, tech giants are collectively spending substantial capital on high-end GPU bulk purchases and building massive-scale data centers. To avoid a sharp surge in current liabilities, they commonly use long-term purchase commitments, sale-leasebacks, and long-term leasing data center financing models. These methods legally take advantage of accounting rules that allow bookings once assets are delivered, pushing forward the pressure of massive capital expenditures—but at the same time, a huge mountain of hidden off-balance-sheet debt accumulates out of sight.

Statistics show that among these five companies, Meta’s hidden debt is the most staggering, estimated at about $420 billion—nearly 2.8 times its on-balance-sheet liabilities. Meta is co-developing a mega data center with Blue Owl Capital (OWL.US) in Louisiana, with total project development costs now soaring from the originally announced $27 billion to over $50 billion. Meta invested only 20% equity, intending to lease the facilities once completed—dramatically reducing its upfront investment on the surface—but reportedly, Meta has agreed to guarantee all losses if the partnership collapses. This potential contingent obligation is not reflected on the balance sheet.

Oracle's hidden debt has exploded more than 30-fold in four years, reaching $273.3 billion. This burden mainly stems from large-scale leasing arrangements, with a significant portion earmarked for the “Stargate” AI data center project in cooperation with OpenAI. Meanwhile, Microsoft, Alphabet, and Amazon’s combined backorders for cloud services and various businesses total approximately $1.45 trillion, seemingly providing future revenue support. However, the ever-increasing deferred payment commitments remain an overhanging shadow.

If Demand Misses Expectations, a “Hidden Bomb” Could Morph into a Liquidity Crisis

Industry concerns focus on the scenario where global AI demand fails to keep pace with widespread high expectations—these deferred debts could suddenly become a real cash flow black hole.

Some technology industry insiders point out that commercial real estate leases at least retain considerable asset value, but AI data centers are packed with rapidly iterating semiconductors, and technological advances cause hardware to depreciate swiftly. If weak AI demand leads to falling rack rates and utilization, tech giants will face substantial impairment losses on these assets.

Even more concerning is the current trend of “circular investment”—tech companies investing in each other's AI services to bolster reported revenue and demand data. In the short term, this masks the uncertainty of end-user demand. However, should real demand fail to absorb the vast capacity under construction, deferred debts will all come due once the new data centers go online—potentially setting off a chain reaction.

Despite tech giants’ continued optimism—Amazon’s AWS CEO and other top management insist that the current investment in AI infrastructure is not speculative—the explosive growth of off-balance-sheet liabilities is now an indisputable reality.

The “Central Bank of Central Banks” Issues a Historic Warning

As tech giants pile on both on- and off-balance-sheet leverage, the Bank for International Settlements (BIS)—often called the “central bank of central banks”—issued a blunt warning in its latest annual economic report. BIS directly likened the current multi-trillion-dollar AI investment frenzy to the 19th-century canal mania, the UK railway bubble, and the dot-com bubble of 2000, pointing out that “these precedents all ended with an investment reversal and economic recession.”

The report estimates that the five U.S. hyperscalers' capital expenditure related to AI from 2025 to 2026 will exceed $1 trillion—beyond what their profits and free cash flow can bear, forcing some companies to take on debt for this purpose.

BIS particularly highlighted the “complex network of private transactions” lurking beneath the surface: Tech giants locking in long-term compute procurement via equity stakes in AI labs, third-party contractors building data centers which are then leased back under long-term contracts with embedded exit clauses. Such disclosure is highly inadequate, and there is a risk of duplicate pledging of a single asset.

Should capital spending come to a sudden halt, the entire supply chain—from chipmakers and infrastructure builders to private and direct investment funds—faces simultaneous revenue disruption. Furthermore, with direct lending exposure to the AI sector quadrupling over five years, and inflation and geopolitical turmoil still high, the financial system could experience even more rapid contagion than a traditional banking crisis.

BIS made it clear in the report: “The scale and speed of the current AI investment boom, coupled with general expectations of significant productivity gains, are highly reminiscent of the aforementioned historical examples. In all these cases, the end result was a sudden reversal of investment leading to economy-wide recession.”