The IMF, the BIS, and the Bank of England have each raised a sharper alarm this month: it isn't inflated AI valuations alone that could crack financial markets, but the borrowed money sitting underneath them.
When Tobias Adrian, the IMF's financial counsellor, outlined the fund's April 2026 Global Financial Stability Report, he didn't spend much time worrying about price-to-earnings ratios. The concern was structural: nonbank financial intermediaries, hedge funds, and leveraged exchange-traded funds had built concentrated AI exposure using borrowed capital, creating conditions where a sentiment shift could force selling that feeds on itself. That distinction matters. Valuations correcting slowly is painful. Forced deleveraging cascades are something else entirely.
The numbers behind the warning are not abstract. According to FINRA data, US margin debt hit a record $1.4 trillion in May 2026, up 54% year on year. Total assets in leveraged ETFs ballooned from roughly $115 billion to $220 billion in under 70 days between late March and early June. On June 5, a 3x leveraged semiconductor ETF dropped 31% in a single session; Nvidia alone shed over $300 billion in market capitalisation during that selloff. Those are not background conditions. They are the amplification machinery the IMF is worried about, already running at speed.
The Bank for International Settlements made the same argument even more starkly in its Annual Economic Report published June 28. The BIS, the Basel-based institution that coordinates the world's central banks, warned that the five largest hyperscalers are set to spend more than a trillion dollars on AI-related capital expenditure across 2025 and 2026, commitments that are already outpacing earnings and free cash flow. The mechanism that makes this particularly dangerous, according to the BIS, is circular financing: hyperscalers take equity stakes in AI labs, which then commit to multi-year purchases of compute from those same hyperscalers. The same asset is effectively pledged multiple times, and the terms of these deals, as the BIS pointedly noted, are typically poorly disclosed. If AI sentiment turns, the interconnectedness of these nonbank channels could make a correction move faster than the 2008 banking crisis did. That's the BIS saying it, not a short-seller newsletter.
On June 30, Bank of England Deputy Governor Sarah Breeden warned at the European Central Bank's forum in Sintra that autonomous AI agents risk triggering market meltdowns if they all respond the same way to the same prompts. The BoE has already begun scenario analysis and simulations focused on herding behaviour: the idea that AI-driven trading systems, trained on similar data and optimised for similar objectives, could synchronise their sell decisions under stress and produce self-reinforcing crashes faster than any human trader could intervene. The BoE's April 2026 Financial Policy Committee record noted that advanced AI is not yet creating systemic instability, but stressed the risk could rise quickly as adoption accelerates. Breeden suggested that existing regulatory frameworks may not be adequate for what's coming.
For founders raising right now, this regulatory convergence has a practical dimension that goes beyond systemic concern. AI startup valuations are currently in the 10x-50x revenue range, with the median hovering around 20x-30x according to Qubit Capital's 2026 analysis. Q1 2026 shattered venture funding records globally at $300 billion, with AI absorbing $242 billion of that, roughly 80% of all VC deployed. Four rounds alone , OpenAI's $122 billion, Anthropic's $30 billion, xAI's $20 billion, and Waymo's $16 billion , accounted for 63% of total global venture activity. Those multiples are not sustained by operating fundamentals alone. They're sustained by the same institutional appetite for AI exposure that now concerns the IMF.
If a forced deleveraging event hits before the next funding cycle closes, the transmission is direct. Institutional investors who are simultaneously running leveraged AI equity positions and sitting on LP commitments to AI-focused funds face a liquidity squeeze that compresses both public and private markets at once. The private credit co-investments that have increasingly filled late-stage AI rounds carry similar exposure: structured on the assumption that AI equity values stay elevated long enough for the credit to be refinanced or repaid. That assumption is now under formal scrutiny from three of the world's most influential financial institutions in the same month.
None of this means the AI investment cycle is over or that the underlying technology doesn't justify substantial capital. The Shiller CAPE ratio sits at 41, near its highest recorded level, but elevated valuations can persist longer than they logically should. The more immediate question is whether the leverage structure underneath the AI trade is as resilient as the narrative around AI itself. Frankly, the IMF, BIS, and BoE are all suggesting it isn't, and they're pointing to the same instruments: leveraged ETFs, hedge fund concentration, nonbank financing chains with limited transparency. If you're a founder with a round due in the next twelve months, the timeline matters now in a way it didn't six months ago. Institutional deleveraging doesn't wait for a convenient quarter.
Also read: The Bank of England is warning that AI trading agents could trigger the next market crisis • Call center giants are being repriced out of existence before AI has finished the job • Ferrari and BMW are ditching copper wiring, and the implications run well beyond the factory floor
