1.65 trillion off-sheet bomb! The “invisible debt” of the big five tech giants has soared eight times in four years. Is the AI arms race likely to spawn the next liquidity disaster?

The Zhitong Finance App learned that the crazy “arms race” of America's top five tech giants on artificial intelligence (AI) infrastructure is spawning a “hidden debt empire” that far exceeds book debt.

Recently, an analysis of the latest financial footnotes from Alphabet (GOOGL.US), Microsoft (MSFT.US), Amazon (AMZN.US), Meta (META.US), and ORCL.US (ORCL.US) shows that the total “hidden debt” accumulated by these five companies through long-term procurement commitments and data center leases has swelled to 1.65 trillion US dollars, which has not only surged eight times over the past four years.

This massive off-balance sheet obligation has made Wall Street and global regulators increasingly worried that once AI demand falls short of expectations, these future payment promises hidden in footnotes may turn into a “debt bomb” that eats up cash flow in a very short time.

Buying GPUs on credit poses hidden debt risks

The so-called “invisible debt” refers to the future payment obligations of an enterprise arising from agreements such as investment, equipment procurement, or long-term leases, but according to current accounting standards, these obligations may not be recognized as liabilities until the assets or facilities are actually received and completed. They neither immediately consume cash on the book, nor do they appear in the main column of the balance sheet. They are only hidden in the notes section at the end of the financial report in the form of footnotes in small print.

Currently, the global battle for AI dominance is heating up, and tech giants are using vast amounts of capital to purchase high-end GPUs in batches and build large-scale data centers. In order to avoid a sharp increase in the size of immediate debt, they generally use long-term procurement commitments and data center financing models such as after-sales leaseback and long-term leasing. This method of operation lawfully and complies with the accounting rules of “to be delivered and then recorded” to move the pressure of huge capital expenses back and forth, but at the same time, it also piled up a huge peak of invisible debt outside the balance sheet.

According to statistics, among these five companies, Meta's hidden debt is the most impressive. It is estimated that it has reached about 420 billion US dollars, which is almost 2.8 times the debt on its statement. Meta is working with Blue Owl Capital (OWL.US) to build a giant data center in Louisiana, and the total development cost of the project has soared from the initial announcement of $27 billion to over $50 billion. Meta only invested 20% of the shares, and when the facility was completed, it was used through leasing. On the surface, the upfront investment was significantly reduced, but according to reports, Meta has agreed to guarantee all losses when the partnership breaks down — this potentially rigid obligation is not reflected in the balance sheet.

Oracle's hidden debt has exploded, expanding more than 30 times over four years to reach 273.3 billion US dollars. This burden is mainly due to large-scale leasing arrangements, and a significant portion of the funding has been earmarked for the “Stargate” AI data center project in collaboration with OpenAI. At the same time, although the total backlog of orders for cloud services and various businesses of Microsoft, Alphabet, and Amazon is about 1.45 trillion US dollars, which seems to support future revenue, the continuous accumulation of forward payment promises still hangs over the head.

If demand falls short of expectations, the “invisible bomb” will turn into a liquidity disaster

The core concern in the industry is that once the growth in global AI demand does not keep up with the industry's high expectations, these deferred debts will suddenly turn into a real cash flow black hole.

Some people in the technology industry pointed out that commercial real estate leasing can still retain considerable asset value, but AI data centers are full of semiconductors that iterate extremely fast, and technological progress has caused hardware to depreciate rapidly. Tech giants will face huge asset depreciation losses if the listing rate and utilization rate decline due to weak AI demand.

What is even more alarming is the currently popular “circular investment” model — technology companies invest in each other's AI services to support book revenue and demand data, which masks the uncertainty of terminal demand in the short term. However, once the actual growth rate of demand is unable to absorb the huge computing power under construction, deferred payment debts will be concentrated when the data center is put into operation.

Although the tech giants themselves are still optimistic, the Amazon AWS CEO and other giant management insist that current investment in AI infrastructure is not speculative, but the explosive growth in off-balance sheet debt is an indisputable fact.

The “central bank of the central bank” issued a historic warning

The Bank for International Settlements (BIS), known as the “central bank of the central bank,” also issued a blunt warning in its latest annual economic report at a time when tech giants are building up their internal and external leverage. BIS directly compared the current trillion-dollar AI investment frenzy to the 19th century canal frenzy, the British railway bubble, and the 2000 internet bubble, pointing out that “these precedents all ended with investment reversal and triggering an economic recession.”

The report estimates that the five largest hyperscaler cloud service providers (Hyperscalers) in the US will total more than 1 trillion US dollars in AI-related capital expenses from 2025 to 2026, which is beyond what their profits and free cash flow can carry, forcing some companies to borrow for this.

In particular, BIS pointed out the “complex private transaction network” hidden behind it: tech giants lock in long-term procurement of their own computing power through equity investment in AI laboratories, and third party contractors build data centers and then rent them back with long-term contracts with embedded exit clauses. The disclosure of these terms is extremely inadequate, and there is a risk of repeated guarantees for the same assets.

Once spending is put on the brakes, the entire supply chain — from chip vendors and infrastructure contractors to private and direct investment funds — will face simultaneous disruptions in revenue streams. In addition, the direct loan sector's exposure to the AI sector has quadrupled in five years. If the impact of the still high level of inflation and geographical conflict is compounded, the financial system is likely to experience a more rapid chain correction than the traditional banking crisis.

BIS clearly stated in the report: “The scale and speed of the current AI investment boom, combined with general expectations of a significant increase in productivity, is highly similar to the historical precedent described above. And the outcome of these precedents was a sudden reversal of investment, which in turn induced an overall economic recession.”