According to a July survey report by Nikkei Asian Review, the hidden debt of five tech giants—Google’s parent company Alphabet, Microsoft, Amazon, Meta, and Oracle—has ballooned eightfold in four years, reaching a staggering $1.65 trillion. That’s well above their combined on-balance-sheet liabilities of $1.35 trillion during the same period.
The report highlights that the red-hot AI market is pushing these companies to keep ramping up capital spending. To quickly build data centers and keep initial costs low, they’re signing long-term leases with data center operators and locking in long-term purchase agreements for GPUs and servers.
Under current U.S. Generally Accepted Accounting Principles (GAAP), undelivered long-term GPU and server contracts, as well as data center leases that haven’t gone live yet, don’t immediately show up on the balance sheet. Companies typically just mention these future obligations in footnotes to their quarterly financial statements. While this practice is technically compliant with accounting rules, it could leave individual investors underestimating the risks.
The scale of current capital spending has blown past the limits of traditional corporate financial models. In fiscal 2025, Alphabet poured $85 billion into capex, mainly to fuel AI computing needs for Google DeepMind. Microsoft spent a whopping $34.9 billion in the first quarter of fiscal 2026 alone, with annual growth expected to be even faster than the previous year. Amazon is planning to invest $200 billion by 2026, with the bulk going to AWS data centers and custom Trainium chip production. Meta expects its full-year 2025 capex to land between $60 billion and $65 billion, focusing on generative AI inference infrastructure. And Oracle, thanks to a potential $300 billion cloud service deal with OpenAI, is projected to see total capex skyrocket to between $90 billion and $95 billion by fiscal 2027—nearly twice its total revenue for fiscal 2025. These aren’t one-time expenses; they’re long-term commitments stretched over years, turning cash flow pressures from a “future risk” into a “present-day burden.”
Where’s the money coming from to support all this massive spending? The answer lies in bond markets and private credit. To keep up, some tech companies are leaning on issuing corporate bonds and new stock to raise funds, fueling an overheated capital investment frenzy.
A February report from Moody’s shows that Amazon, Meta, Alphabet, Microsoft, and Oracle—these five hyperscale data center operators—have signed but not yet booked long-term data center lease commitments totaling roughly $662 billion. That’s equal to 113% of their combined adjusted debt. As these leases kick in over the next few years, more than $500 billion in data center debt will eventually hit their balance sheets, potentially sending adjusted debt levels through the roof. If AI revenue growth can’t keep pace with the debt load, it’ll directly squeeze financial flexibility. Moody’s analyst David Gonzales warns: “This unrecorded debt burden creates a risk profile far higher than what traditional financial statements reveal.”
Beyond the hidden liabilities from off-balance-sheet contracts, the tech giants’ aggressive expansion is also brewing a double whammy of “shadow lending” and “circular investment” risks, steadily inflating the AI investment bubble. To keep up with rising AI capex, many tech firms are increasingly issuing corporate bonds and new shares, raising funds from the secondary market to fuel expansion and driving up industry investment heat.
In March, the Bank for International Settlements published a special report, labeling this model—where companies avoid adding visible debt on the books while relying on institutional fundraising and off-balance-sheet contracts—as classic “shadow lending.” While it might make near-term financials look better, it actually stretches out long-term fixed financial burdens and hides persistent cash flow risks.
Meanwhile, the media has also revealed a unique circular investment risk in the AI space. Hardware giants like Nvidia and U.S. tech titans form a closed-loop cash flow: tech companies inject capital into data centers and AI operators, and that money flows right back as revenue from GPU purchases and cloud service payments. This cycle artificially pumps up industry sentiment, masking real consumer demand at the end of the line. It directly fuels the risk of global overinvestment in data centers and excess computing capacity.
Still, the tech giants remain optimistic about their ability to pay off the debt. As of the end of March, Microsoft, Alphabet, and Amazon reported a combined $1.45 trillion in hand orders for cloud services and other businesses. They believe future revenue will easily cover off-balance-sheet debts. AWS CEO Matt Garman publicly stated that Amazon’s investments “aren’t speculative,” but rating agencies like Moody’s are clearly taking a more cautious stance.
Facing this trend, capital markets are shifting from enthusiastic support to wariness. Goldman Sachs, in its latest research report, admitted it had underestimated the resilience of cloud providers’ capex. It has hiked its 2026 global AI capex forecast from $465 billion to a massive $765 billion, explicitly advising investors to “go long on hyperscale cloud providers and underweight semiconductors.” The bank argues that cloud platforms hold the distribution power and data loops in the AI ecosystem, making their long-term value irreplaceable even if short-term ROI looks weak.
