Why the BIS sees the AI capex boom as a debt-market risk

In its annual report, the Bank for International Settlements recast the AI investment spree as something broader than a pure equity-multiple story. Debt-funded spending, opaque financing structures and wider fiscal fragilities, the BIS said, are starting to reinforce one another while investors still treat AI demand as a one-way trade.

Scale is what moves the warning beyond a handful of mega-cap stocks. The five largest hyperscalers are on track to spend more than $1 trillion across 2025 and 2026 on AI-related capital expenditure, according to The Register’s summary of BIS estimates. CNBC reported this week that Amazon, Microsoft, Meta and Google alone are planning roughly $700 billion of 2026 capex for AI infrastructure. Once spending reaches that scale, the question for central bankers stops being whether Nvidia or another supplier can justify its valuation. It becomes what happens if the financing chain behind the buildout wobbles.

BIS officials are not calling AI investment inherently destabilising. The report credits the boom with supporting growth and loosening financial conditions. Concentration is the concern. When one investment theme props up growth, credit creation and equity sentiment at the same time, any reversal travels through more channels than a simple correction in software stocks.

“Policy actions must reinforce each other to avoid a pull and push on the global economy. Ultimately, success depends on sound fiscal and financial foundations.”

Pablo Hernandez de Cos, via Reuters

MarketWatch framed the issue as something larger than frothy valuations. Its read of the BIS warning put the bigger vulnerability in the debt issuance of hyperscalers, AI labs and the engineering contractors building out the physical stack. Equity investors may have paid too much for future AI profits; lenders, private-credit funds and infrastructure vehicles have also begun underwriting those profits as if the demand curve were already settled.

Where the borrowing sits

The capital stack around AI is sprawling enough that stress may surface far from the companies investors associate with the trade. Data-centre developers, contractors, utilities, chip suppliers and private-credit backers all sit downstream of the hyperscaler spend. CNBC’s reporting on Digital Realty and Blackstone’s Virginia data-centre deal offered a concrete example: real-estate and infrastructure capital is being pulled into AI buildout, and pricing has started to reflect how crowded the rush has become.

For the BIS, opacity is the recurring concern. The classic late-cycle problem is not borrowing by itself; it is borrowing distributed across entities with different disclosure standards and different assumptions about exit liquidity. Reuters reported that the BIS highlighted opaque financing arrangements around the AI boom. MarketWatch went further, pointing to the possibility that a U.S.-led repricing could morph into a broader credit crunch if lenders and investors discover too much of the sector has been financed on similar optimistic assumptions.

Borrowed money has also spilled outside corporate balance sheets. Semafor reported that $1.4 trillion in margin debt helped intensify stock swings tied to AI-related bets, with spillovers from South Korea into U.S. trading last week. That is a different transmission channel, but it lands in the same place: an investment theme that began as a growth story is being financed and traded with enough borrowed money to matter for wider market plumbing.

AI may be generating a familiar kind of fragility through unfamiliar assets. In the 2000 dotcom cycle, the obvious excess sat in listed technology stocks. In the pre-2008 cycle, it sat in credit structures whose risks were assumed away until funding dried up. The current AI buildout touches both worlds. It depends on equity exuberance to sustain valuations, but it also depends on debt markets, private capital and long-duration infrastructure bets to keep physical capacity expanding.

“The new fiscal-financial stability nexus may mean more frequent and sharper drops in sovereign bond values.”

Frank Smets, via Reuters

Smets’s point matters because sovereign-bond volatility raises the cost of capital just as governments and companies alike are trying to finance huge investment programmes. That makes the AI boom more sensitive to rate shocks, and it makes public debt dynamics more sensitive to any disappointment in the productivity gains that are supposed to justify the spending.

Why a capex slowdown would spread quickly

Productivity is the bullish answer to the BIS warning. If generative AI lifts output, today’s buildout could look like rational front-loaded investment rather than reckless overspending. The BIS does not deny that possibility. It questions the market’s assumption that the spend can keep rising without interruption and without broader consequences if it stalls.

Axios argued that the historical warning in the BIS report is about dependence: the global economy is leaning on one investment boom more than many investors care to admit. Unpacked, the trade is a chain. Hyperscaler orders support chip demand; chip profits feed equity indices. The buildout also supports landlords, utilities, financing vehicles and private-credit lenders. If the expected payoff from AI systems slips, the first pain may show up in spending plans, not in a spectacular technology failure.

A capex retrenchment would not only squeeze the companies ordering servers. It would also hit the firms that borrowed to build facilities, the creditors that assumed continued utilisation growth, and the households whose retirement accounts have become more exposed to a narrow cluster of AI beneficiaries. MarketWatch estimated that as much as 15 per cent of some portfolios is exposed to AI-sector lending and related assets, a sign that the boom has travelled well beyond equity benchmarks.

Another MarketWatch analysis noted that if analyst forecasts for a sharp deceleration in hyperscaler capex from 2027 prove right, the semiconductor trade could come under severe pressure and induce a more sustained correction in both equity and debt markets. That is roughly the scenario the BIS wants policy makers to contemplate now, before the buildout becomes even more entangled with pension capital, sovereign balance sheets and private-credit products sold as insulated from public-market volatility.

Why central bankers are uneasy

Regulators do not want to look anti-growth at a moment when AI may yet deliver real productivity gains. They also do not want a repeat of the pattern in which innovation is celebrated while the financing behind it escapes scrutiny. The BIS’s annual report lands in that gap: it praises the upside, then warns that fiscal positions, bond-market volatility and financial intermediation are not built to absorb endless optimism.

Those concerns are spreading beyond Basel. CNBC reported on Thursday that European bankers and regulators are warning AI is already outpacing the rulebook, particularly in a financial system with fewer market-based funding channels than the United States. The BIS reaches a similar conclusion in softer language: it says it would be unwise to be overly prescriptive today, which is another way of admitting that policy is still trying to map a boom private capital has already scaled.

Fiscal pressure adds another layer. The BIS has spent much of the past year warning about the interaction between high public debt and more volatile bond markets. Add an AI buildout that increasingly relies on debt, private credit and infrastructure vehicles, and the system becomes more reflexive. Rising yields push up financing costs. Higher financing costs make investment more brittle. A weaker investment outlook then weighs on the growth narrative that had been helping markets tolerate the debt in the first place.

Read the BIS warning less as a call that the AI bubble pops tomorrow and more as a signal about where the story has moved. The next phase of the AI trade will be judged less by excitement around models than by the resilience of the balance sheets underwriting them. When central bankers start treating AI capex as a financial-stability question, they are signalling that the story has moved out of Silicon Valley and into the credit cycle.