Pablo Hernandez, president of the Bank for International Settlements (BIS), warned in a public speech on Thursday that the global artificial intelligence investment boom is being driven by opaque debt financing and private credit. If future commercial returns fail to meet market expectations, he said, the result could pose a real threat to financial stability.
Hernandez stopped short of saying an AI bubble is certain to burst. Still, he said: “I am not saying this is the inevitable outcome of the AI boom, but the scale and speed of the current investment surge, as well as market expectations for commercial returns, do call for caution.”
Big Tech plans more than $1 trillion in AI spending across 2025 and 2026
Using the latest data cited by the BIS, Hernandez laid out how quickly AI-related investment is expanding worldwide.
- Five major technology companies — Microsoft, Alphabet, Meta, Amazon and Apple — plan to spend more than $1 trillion on AI-related projects in 2025 and 2026.
- A Bridgewater report said Microsoft, Alphabet, Meta and Amazon together carry a market value of about $12 trillion, and their AI infrastructure spending alone will reach $650 billion this year.
- Global AI investment now stands at about $500 billion and is projected to rise to $3 trillion to $4 trillion by 2030.
Hernandez said capital expenditure by the largest AI companies has already moved beyond their cash flow, with the gap financed through debt and private credit. He described that as an investment arms race among industry leaders.
He compared the current AI boom with four earlier technology manias
In the speech, the BIS chief reviewed four historical episodes in which a technological breakthrough triggered a rush of capital, only for the pattern to end in painful correction.
- Canal Mania in the 1830s, when British capital poured into canal construction and many projects later failed to generate enough traffic.
- Railway Mania in Britain in the 1840s, when railway shares climbed to extreme valuations and the collapse was followed by widespread bank failures.
- The electricity revolution in the 1920s, when electrification did improve productivity but the initial investment wave far exceeded the economy’s ability to generate returns.
- The late-1990s dot-com bubble, when .com company valuations detached from reality and the 2000 crash wiped out trillions of dollars in equity value.
“Each time, more capital was drawn in than the eventual returns could justify. Each subsequent correction had economy-wide effects,” Hernandez said.
He also argued that the problem is not the technology itself. Railways, electricity and the internet all became core infrastructure for the modern economy. AI, he said, has real potential as well. The BIS acknowledged measurable productivity gains from AI in software development, consulting and professional writing. The issue, in his view, is whether the scale and speed of investment have outpaced what the economy can absorb.
Opaque links tie together chipmakers, cloud providers and AI startups
Another central warning from the BIS involved the hidden structure of the financing chain. Chipmakers such as NVIDIA, AMD and Broadcom, hyperscalers including Microsoft Azure, AWS and Google Cloud, and AI startups are linked through complex financing arrangements that are difficult to price and lack transparency.
- If profit expectations for AI companies fall short, pressure could move backward through the capital chain to chip suppliers and cloud providers.
- The size and concentration of the private credit market are not fully visible to regulators, making risk accumulation harder to detect in real time.
- US stocks account for an outsized share of global equity markets, so a concentrated sell-off in AI-related names could spread internationally through portfolios and asset-management channels.
Hernandez added: “In some jurisdictions, the large revenues generated by AI-related exports may also fuel domestic asset bubbles and worsen financial stability risks.”
The speech also pointed to implications for Taiwan and Asia
The report tied the BIS warning to Taiwan and the wider Asian economy. Taiwan sits at the center of the global AI hardware supply chain, with TSMC, UMC, Delta Electronics, Quanta and Wistron deeply involved in supplying AI chips and servers to NVIDIA and hyperscalers.
- If AI capital expenditure growth slows, orders across Taiwan’s related supply chain could contract as well.
- If US technology stocks undergo a sharp correction, the move could reach Taiwan equities through foreign holdings and market transmission.
- If global AI investment expectations fall back from $4 trillion to a more conservative $2 trillion to $3 trillion, supply-chain capital expenditure plans could be reassessed.
The report set that against the 2022 technology stock correction and the 2023 rally driven by NVIDIA earnings, and said investors and policymakers in Taiwan should watch whether the “investment arms race” identified by the BIS is moving toward an unsustainable stage.
Central bank mandates stay the same, but monitoring gets harder
Hernandez said AI will not change the core mission of central banks, which remains price stability and financial oversight. What may change is how hard the economy becomes to interpret. Automated trading, AI-driven shifts in consumer behavior and distortions in productivity data could all make traditional indicators less sensitive.
He said the eventual economic effect of AI will depend on two key variables: whether the gains are shared broadly across industries and social groups, and whether policymakers are willing to invest at the same time in skills training, infrastructure and competition policy.