Amazon, Alphabet, Microsoft, Meta and Oracle — the five largest technology companies in the world — plan to spend over one trillion dollars on AI infrastructure alone between 2025 and 2026. The Bank for International Settlements (BIS), whose recommendations are read by the central banks of 60 countries, speaks directly: the pace of spending is outpacing actual profits.
But the real problem that the BIS identified in its annual report is not the size of the spending, but how it is being financed.
Debt that doesn't appear on the balance sheet
Hyperscalers issued over 100 billion dollars in corporate bonds in 2025, and most of this debt is long-term, maturing in over five years. This is a bet that AI will pay for itself — but when exactly, nobody knows.
The BIS identified something more alarming: companies are actively using off-balance-sheet structures — special purpose vehicles, joint ventures and private credit fund agreements — that hide the real scale of indebtedness. The BIS describes this as "shadow borrowing."
These schemes create new networks of financial ties between technology companies, banks and private credit markets. Traditional balance sheet analysis no longer provides the full picture.
"A reversal of optimism about AI could have serious financial consequences given the growing leverage of AI companies and their impact on credit markets."
— BIS Annual Report, 2026
Circular financing: money moving in circles
The BIS separately highlights so-called circular financing: hyperscalers participate in the equity capital of AI laboratories, which then commit to buying chips or computing power from them. This means that revenue is partly generated by their own investments — and looks better from the outside than it actually is.
According to analysts cited by the BIS, enthusiasm is also fueled by IPOs: the recent SpaceX exit and planned public offerings by Anthropic and OpenAI. Some market observers are already comparing the situation to the electrification boom of the 1920s and the dot-com bubble of the late 1990s.
Why the crisis could extend beyond the tech sector
The BIS describes AI as a general-purpose technology — on the level of electricity or the internet. This means that systemic risks are not localized.
According to LVRG Research analyst Nick Rack, AI sector financing relies on massive debt and highly leveraged non-bank structures that can quickly unravel and amplify a crisis.
Investors assessing hyperscalers by traditional balance sheet metrics risk underestimating real leverage: shadow structures create a gap between declared and actual debt.
- If AI doesn't deliver expected profits — financing will be slashed sharply.
- If major bond holders begin to sell off — pressure on all credit markets will increase.
- If correction affects US stocks — the wealth effect will hit consumer spending harder than in previous cycles, due to US dominance in global portfolios.
However, for now the global economy is demonstrating "remarkable resilience" in 2025, partly thanks precisely to AI investments, which are sustaining confidence and mitigating the impact of trade tariffs and geopolitical uncertainty.
The question is not whether correction will happen — but when the first commercial results from AI become significant enough to justify the trillion dollars in commitments. If OpenAI and Anthropic's IPOs happen before the market sees real profits — that will be the moment when it becomes clear whether enthusiasm has turned into a bubble.
