Off-Balance-Sheet Liabilities of America's Big Five Tech — The $1.65 Trillion Shadow Debt in AI Infrastructure Investment

In late July 2026, Nikkei completed a rather unusual "ledger-flipping" investigation: it analyzed the latest financial statements of five American tech giants — Alphabet, Microsoft, Amazon, Meta and Oracle — in search of a number that does not appear on the balance sheet.

The result is startling. The combined "hidden debt" of these five companies in their most recent quarter is estimated at $1.65 trillion — more than the roughly $1.35 trillion of on-balance-sheet debt they carry in total.

This number is "hidden" not because it has been concealed, but because accounting standards permit these liabilities to stay off the balance sheet.

How the Debt Is "Hidden"

The premise for understanding this pile of debt is understanding how AI data centers are built.

Big Tech is not constructing its own data centers. Instead it signs long-term lease contracts with other data-center operators — the operators provide the land, the buildings and the power infrastructure, and the tech companies move in as tenants. Under current accounting standards, so long as a data center has not yet been put into operation and its GPUs and servers have not yet been delivered, these long-term purchase and lease contracts can be treated as "off-balance-sheet items" and kept off the balance sheet.

Concretely, a tech company only books an expense once it actually draws on the portion of computing resources it has committed to. But the long-term obligations it has signed — Meta's $50 billion AI data center, for instance, or the hundreds of billions of dollars in GPU leases under the Oracle-led "Stargate" project — exist only as footnotes to the financial statements, buried in more than a hundred pages of fine print that most individual investors will never read.

Meta's problem is especially acute. Its off-balance-sheet debt stands at roughly $420 billion — 2.8 times its on-balance-sheet debt.

Whose "Technological" Leap — and Whose "Bill-Paying" Anxiety

Zooming out, this is not simply a question of a financial figure. Nested inside it are at least three layers of structural contradiction.

Layer 1: The Clash Between the Technology Race and Accounting Conservatism

The essence of accounting standards is conservative — they tend to "wait until something has happened before booking it." The essence of the AI investment race, by contrast, is aggressive — every participant is betting that demand will explode over the next three to five years, so capacity must be locked in now. Lease contracts sit naturally at the intersection of the two: they are already legally irrevocable commitments, yet accounting standards allow them not to be recognized as debt in economic terms. The result is that the financial statements say "everything is normal," while the actual locking-up of resources is already accelerating the consumption of future freedom of action.

Layer 2: The "Circular Investment" Problem in the AI Industry

The article reveals a structure that has rarely been mentioned in discussions of Nvidia's growth: current AI demand is, to some degree, self-created through "circular investment." Nvidia and other tech giants invest in data-center operators and AI startups, and those investments then flow back to them in the form of GPU purchases and cloud-service fees. This closely resembles the maneuver of the dot-com bubble era, when telecom-equipment makers financed their emerging internet customers — a practice that also triggered the question of "real demand or manufactured demand."

Nikkei's reporting does not declare this a bubble, but economists at the Bank for International Settlements had already sounded a warning in a March report, calling it "shadow borrowing" — a mechanism for raising funds from institutional investors without adding to on-balance-sheet liabilities.

Layer 3: "The Bigger Risk Is Delaying Investment" — and the Market Logic of "Slow Down and You're Out"

One of Zuckerberg's lines, quoted in the report, captures the competitive structure of the current AI investment boom with precision: "The bigger risk is delaying investment." At a stage when the technology path has not yet converged and business models have not yet been validated, abandoning investment means voluntarily leaving the table. But for tech giants accustomed to the "asset-light, high-margin" profile of software companies, this structural transformation of the balance sheet into something "heavier with every signature" — even if the liabilities sit off the books — is itself a quiet revolution.

Enron's Cautionary Precedent

The report does not forget to cite a precedent at the end. An executive at an audit firm told Nikkei: "In fact, there is growing concern that the real financial burden on the books of these companies is far greater than what the balance sheet shows." For an auditor to say such a thing is itself a kind of signal. The article goes on to invoke Enron — which filed for bankruptcy protection in 2001 after hiding off-balance-sheet debt — even as it hastens to add that "the nature of the companies is different."

Whether the analogy is apt remains to be verified, but it at least raises a proposition that deserves to be taken seriously: when the scale of non-debt commitments exceeds the scale of debt commitments, and the former need not be reflected in mainstream financial statements, the market's judgment of a company's true leverage depends entirely on whether it is willing to turn to the footnotes of those statements.

Morgan Stanley has already analyzed this in detail in an investor report. Moody's, too, pointed out in a February report that the value of lease contracts tied to data centers not yet in operation is ballooning.

Nvidia's Circular Financing — From Chip Supplier to Capital Intermediary

On July 27, 2026, as the market was still digesting Nikkei's off-balance-sheet investigation, Bloomberg sent a different kind of signal: Nvidia is pushing forward a series of AI-related deals totaling more than $750 billion, and these massive investments are rekindling market concerns about "circular financing."

Circular financing refers to a structure in which companies in an industry chain invest in, take equity stakes in, or provide financing to one another, thereby generating stronger demand and higher valuations — but that demand may not entirely come from the real market.

Nvidia's recent moves illustrate the mechanism clearly.

The $500 Billion Collaboration with SK Group

Nvidia and South Korea's SK Group announced an AI business collaboration worth more than $500 billion. SK Group is simultaneously a major customer of Nvidia and a core partner in HBM memory — a cross-holding and order relationship that makes it increasingly difficult to distinguish "real demand" from "circularly generated demand."

OpenAI's $250 Billion Financing Arrangement

Nvidia is in discussions to provide OpenAI with roughly $250 billion in financing to help it lease US data-center compute capacity. Even more striking, Nvidia may also arrange about $350 billion in financing for OpenAI specifically to purchase Nvidia chips. This means Nvidia participates in the same transaction in two roles: as both capital provider and chip supplier.

SSI's $5 Billion Investment

Nvidia has also invested roughly $5 billion in Safe Superintelligence Inc. (SSI), the AI company founded by former OpenAI chief scientist Ilya Sutskever. The common thread across these investments is that Nvidia is no longer merely an AI chip supplier — it is becoming a major capital participant in the entire AI infrastructure ecosystem.

Over the past few months, a number of investors — including Goldman Sachs and "Big Short" Michael Burry — have been warning about the dangers of this pattern. Burry posted on social media, citing Bloomberg's report: if AI application profitability does not meet expectations, these highly interlinked financing chains could become a source of systemic risk for the entire industry.

How Circular Financing Changes the Risk Picture

The most immediate impact of this finding is that it turns the static picture Nikkei painted of "$1.65 trillion in off-balance-sheet debt" into a dynamic one.

Nikkei's reporting had already revealed a structural transformation of the balance sheet — the Big Five were locking in massive future commitments through off-balance-sheet leases. Bloomberg's reporting now shows that the locking-in process itself is being actively amplified by Nvidia's own financing behavior: Nvidia is not passively waiting for customers to place orders with their own capital; it is actively helping customers obtain the funding to buy its own chips.

Investors' assessment of risk has already begun to show in the stock price. On July 27, Nvidia's shares fell about 5 percent, their biggest one-day drop since June 5. Apple, at a market capitalization of nearly $5 trillion, overtook Nvidia to become the world's most valuable company again.

Whether that price move constitutes direct evidence of "the market pricing in circular financing risk" is still too early to say — the Philadelphia Semiconductor Index fell more than 2 percent overall on the same day, with SK Hynix and Micron also down in tandem. But the view of Bloomberg analyst Mandeep Singh is worth noting: "Nvidia wants to make sure AI infrastructure build-out continues at its current pace, but that is also the biggest risk it faces. If the pace of construction pauses, even for six months, it could have a serious impact on the company."

This also raises the question that Nikkei's report left at the end without fully exploring: when the "risk of delaying investment" is greater than the "risk of over-investment," is the entire industry operating in a mode of rationally moving toward the irrational?

The Wall Street Journal's $7 Trillion Warning — From "Shadow Debt" to "Bubble Burst"

On August 2, The Wall Street Journal delivered the most blunt blow yet to the issue this page has been tracking: investment in artificial intelligence — above all data-center construction — may be manufacturing a financial bubble, and when it bursts it will "drag the whole market down with it." The article's arithmetic is strikingly concrete: spending on data-center construction over the next four years is estimated at up to $7 trillion, and if productivity growth cannot match those costs, that alone would be enough to seriously damage the economy.

This figure sits on a different order of magnitude from the $1.65 trillion in off-balance-sheet debt that Nikkei's calculation at the top of this page arrived at — $1.65 trillion is the shadow borrowing the Big Five have already tucked away on their books; $7 trillion is the entire industry's capital-expenditure expectation for the next four years. The former answers "how much has already been owed"; the latter answers "how much more is about to be poured in."

The WSJ singles out in particular a shift in the financing structure: the scale of AI investment made with borrowed money is rising. That means the shock of a bursting bubble would no longer be confined to paper losses for equity investors — it would propagate along credit chains into the financial system itself, "by no means something the market can ignore." A report by US news outlet NOTUS in early July corroborates this: in-service analysts at the US Treasury have already warned that the American economy faces high risk if the AI bubble collapses.

Who Is Making Money in the AI Economy — The Four Variables of Winning and the Logic of Bottlenecks

In the early hours of August 3, Lingshi Xiantan (a Chinese commentary account) passed along an analysis of the winners of the AI economy. Its core judgment, in the briefest terms, is this: the contest for AI dominance is often reduced to a two-horse race between the United States and China, but the real list of winners is much longer — and who wins depends on four things: labor structure, fiscal capacity, social readiness, and supply-chain relationships.

This article sits at the opposite end of the same topic this page has been tracking through the lens of debt. Nikkei calculated "how much the Big Five have already borrowed"; the WSJ asked "will the bubble burst?"; this one inverts the question — even if it does not burst, into whose hands will the money flow?

Among the answers the article offers is one counterintuitive conclusion: the United States and China lead in research and development, but each carries its own vulnerabilities; conversely, countries that hold chips, critical minerals or energy — even if they cannot become all-around winners — can preserve geoeconomic influence through a "bottleneck" advantage. The holders of chips, minerals and energy do not need to win the entire race; they only need to make sure that the contestants cannot do without them.

This information does not conflict with the evidence chain on this page — it completes the other half of the picture: the distribution of returns. The bubble narrative concerns the risk side — leverage, borrowing, off-balance-sheet debt; the bottleneck narrative concerns the returns side — even if the bubble does not burst, who takes the profits. Only together do the two lines form the full picture: America's tech giants are shouldering the risk, while countries holding critical resources are locking in the returns.

The Free-Cash-Flow Cliff — The Second Ledger of the Big Five's AI Bet

What this page had previously calculated was the stock side: $1.65 trillion in shadow debt. In the early hours of August 4, The Washington Post's analysis of S&P Global data flipped the ledger over to the flow side: free cash flow at the five leading AI companies — Amazon, Google, Microsoft, Meta and Oracle — is projected to shrink to nearly zero over the coming year, falling further to negative $125 billion in 2027.

Two data points make this table uglier than it was last week. The first is that Amazon has raised its AI data-center spending plan, widening the hole on the books accordingly; the second is that over the past three months, Amazon and Google have burned more cash than any other large American company — what S&P Global calls an "embarrassing milestone." Elon Musk's SpaceX is expected to post an even worse cash burn this week, driven by the same cause: AI investment.

Put the stock ledger and the flow ledger together, and the evidence chain on this page closes: the off-balance-sheet liabilities are what has already been owed; free cash flow trending toward zero is what is happening now. Nikkei calculated "how much has been borrowed"; S&P calculated "how much is left." The two ledgers point in the same direction — capital expenditure on AI infrastructure is squeezing every last ounce of elasticity out of the income statement. The Washington Post also identifies the transmission end of this chain: AI spending is pushing up electricity prices and consumer-electronics prices, while Americans' retirement savings are deeply entangled with the stock market — and the market is growing anxious that the AI boom may prove fleeting.