Introduction

On Sunday, June 28, 2026, the Bank for International Settlements published its Annual Economic Report and used it to deliver one of the starkest warnings any major financial institution has issued about artificial intelligence to date. An AI bust, the report stated, ranks alongside inflation and fiscal stress as among the most alarming threats to global prosperity at present. The institution that coordinates the world's central banks did not frame this as speculation about a distant possibility. It described an AI bust as a live "pressure point" that "demands attention," sitting on top of financial vulnerabilities that, in the BIS's own words, could amplify any shock that arrives.

The specific phrasing matters. BIS General Manager Pablo Hernandez de Cos described competitive pressure between technology companies racing to secure market share in AI as having potentially driven investment beyond levels that realistic returns can justify. "Disappointment in returns could trigger a sudden pullback in financing and turn the capex boom into a protracted investment bust, with potential knock-on effects on financial conditions," the report stated. That sentence describes something considerably more serious than a stock market correction. It describes a mechanism by which an AI valuation reset could travel out of equity markets and into the credit system — the web of loans, bonds, and debt obligations that underpins the broader economy. Understanding how that mechanism works, and why the BIS is the institution sounding the alarm, is the most important financial story of the year that most people have not yet fully absorbed.

What the BIS Is and Why Its Warnings Carry Weight

The Bank for International Settlements, based in Basel, Switzerland, is often described as the central bank for central banks. It does not set interest rates or issue currency. Its function is to coordinate, research, and advise the world's monetary authorities — the Federal Reserve, the European Central Bank, the Bank of Japan, and dozens of others — on the risks facing the global financial system. Its annual economic report is one of the most closely watched documents in international finance precisely because the BIS has no domestic political constituency to please and no market position to defend. When it identifies a risk, it is making a technical judgment rather than a political or commercial one.

The institution's credibility on exactly this kind of warning is well established historically. As the BIS's own economists William White and Claudio Borio demonstrated in the years before the 2008 financial crisis, the organisation was among the earliest and most consistent voices flagging dangerously elevated housing prices and the leverage building up underneath them — warnings that were largely dismissed at the time by figures including former Federal Reserve Chair Alan Greenspan, and that proved prescient when the crisis arrived. That track record is the reason the 2026 warning has generated the level of attention it has. The BIS does not issue alarmist statements casually, and its specific language — describing the current moment as one where "the global economy remains caught in the crosscurrents of progress and peril" and where "resilience is being increasingly tested and strained" — reflects a institution choosing its words with the deliberate caution that characterises central banking communication.

The Scale of What the BIS Is Actually Describing

The numbers underlying the BIS warning are what give it weight beyond rhetoric. The five largest hyperscale cloud computing companies — Alphabet, Amazon, Meta, Microsoft, and Oracle — are on track to commit more than $1 trillion to AI-related capital expenditure across 2025 and 2026 combined. That spending pace is outstripping the earnings and free cash flow these companies generate, which is forcing a structural shift in how the AI buildout is being financed. Companies that historically funded their infrastructure investment from their own operating cash flows are increasingly turning to debt markets instead, exploiting corporate credit spreads that sit near century lows to secure financing quickly and cheaply.

The BIS's companion research, published as BIS Bulletin No. 120, drills into the specific mechanics of that financing shift in granular detail. Private credit funds — lenders operating outside the traditional banking system, with considerably less regulatory oversight — originated more than $40 billion in loans to AI-related companies in 2025, compared with roughly $3 billion in 2010. AI-related direct loans now account for a meaningful share of the entire private credit market. This matters because private credit and other non-bank financial intermediaries, including hedge funds and specialised lending vehicles, have become deeply embedded in how the AI buildout is funded, and these institutions operate with less transparency and less regulatory scrutiny than the commercial banks that financed previous infrastructure booms.

The report also identifies what it calls circular financing as a specific structural vulnerability. The pattern works like this: a hyperscaler takes an equity stake in an AI lab. That AI lab, in turn, commits to multi-year purchases of computing power or chips from the same hyperscaler or its partners. The same dollars effectively move in a loop between a small number of interconnected companies, with each transaction simultaneously functioning as revenue, as investment, and as a contractual commitment that gets counted as backlog or future earnings. The BIS describes these circular deals as opaque arrangements where the underlying assets may, in effect, be pledged multiple times across different parts of the financing structure — a description that should sound familiar to anyone who studied the layered, hard-to-untangle financial instruments that obscured risk in the run-up to 2008.

How a Valuation Correction Could Spread Into Credit Markets

The mechanism the BIS is describing has several connected stages, and understanding each one clarifies why a correction in AI company valuations could become considerably more dangerous than a simple stock market decline.

The first stage is a disappointment in returns. If AI companies, after years of extraordinary capital expenditure, fail to generate revenue and profit growth that justifies the scale of investment that has gone into building data centres, training infrastructure, and chip supply chains, the market's confidence in the sector's trajectory weakens. This is the scenario BIS economists modelled using contest-theory analysis: as competitive pressure drives capital expenditure higher across multiple companies racing for market share simultaneously, the net economic surplus for the sector as a whole — total payoffs minus total investment costs — can decline and, in adverse scenarios, turn negative even while individual companies post impressive headline revenue figures.

The second stage is where the danger compounds. A disappointment in returns does not stay confined to equity valuations. Because so much of the AI buildout has been financed through debt rather than equity, and because that debt increasingly flows through private credit funds and other non-bank lenders rather than traditional banks, a pullback in investor confidence can trigger what the BIS describes as a sudden pullback in financing. Companies that assumed continuous access to cheap credit to fund the next phase of data centre construction find that access constricting precisely when their revenue trajectory is also disappointing expectations. The capex boom that depended on continuous financing availability can convert quickly into what the BIS calls a protracted investment bust.

The third stage is the one that distinguishes this risk from a conventional market correction: interconnectedness and speed. Zhang Tao, the BIS's chief representative for Asia and Pacific regions, made the point directly: "If the market has any sort of correction, the interconnectedness of the financial system and interplay of vulnerabilities could mean the speed of a correction could be much faster than previous banking crisis episodes." Because circular financing arrangements link hyperscalers, chipmakers, and AI labs together through equity stakes, multi-year purchase commitments, and debt obligations that all depend on the same underlying assumption of continued AI revenue growth, a shock to any one part of that web can propagate through the others faster than a traditional credit crisis would, because the linkages are denser and the assets involved have been valued, and in some cases pledged, multiple times across different parts of the same interconnected structure.

How This Compares to the Dot-Com Crash

The comparison to the dot-com collapse of 2000 is the one most people reach for instinctively, and the BIS report itself situates the current AI boom in a longer historical lineage that includes the dot-com era alongside the canal mania of the 1830s and the British railway bubble of the 1840s. Each of those episodes, as Fortune's analysis of the BIS report observed, began with a genuine technological breakthrough that attracted more capital than commercial returns could ultimately justify, and each ended in a recession.

The comparison is useful but incomplete in one critical respect. The dot-com bust, severe as it was for technology stock valuations and for the venture capital industry, did not produce a systemic credit crisis on the scale of 2008, in large part because the bubble was financed predominantly through equity rather than debt. Companies burned through venture capital and public market valuations collapsed, but the broader banking system did not face the kind of leveraged, interconnected exposure that turned the 2008 housing collapse into a global financial crisis. The current AI buildout looks structurally different on exactly this dimension. The shift toward debt financing — including the unprecedented growth in private credit lending to AI companies — and the prevalence of circular financing arrangements means that an AI bust in 2026 carries credit system risk that the dot-com crash largely did not. The BIS's own framing makes this comparison explicit: the circular financing it describes is, in its assessment, more opaque than the layered investment trusts of the 1920s that amplified the severity of the Great Depression, even though the underlying technology driving the current boom is, unlike many 1920s speculative vehicles, genuinely transformative.

What Early Warning Signs Are Being Watched

Several specific indicators are the ones that investors, regulators, and the BIS itself are watching most closely for signs that the risk described in the report is beginning to materialise. The first is corporate credit spreads for technology and AI-related debt issuance: spreads that remain compressed near historic lows reflect continued investor confidence, while any sustained widening would signal that the market is beginning to price in the disappointment-in-returns scenario the BIS describes. The second is the pace of debt issuance by hyperscalers specifically — a continued acceleration in borrowing to fund capital expenditure, even as revenue growth shows signs of decelerating relative to spending, would be a warning sign that the gap between investment and realistic returns is widening rather than closing.

The third indicator is the scale and growth rate of private credit exposure to AI companies, which the BIS Bulletin No. 120 data already shows expanding more than tenfold in fifteen years. Continued rapid growth in this less-regulated lending channel, without commensurate transparency about underlying loan quality and collateral structures, would represent exactly the kind of blind spot the BIS warns regulators currently have limited visibility into. The fourth is the behaviour of non-bank financial institutions more broadly: the BIS report separately notes that non-bank financial institutions have become the largest holders of sovereign debt in advanced economies, with their share rising from 44% in 2021 to 53% in 2025, and that these leveraged, funding-dependent institutions can amplify liquidity shocks through repo and derivative markets in ways that traditional bank balance sheets historically did not. A stress event in sovereign bond markets, interacting with AI-related credit stress simultaneously, is precisely the kind of compounding scenario the BIS report is structured to highlight as a connected risk rather than an isolated one.

What a Responsible Response Looks Like

The BIS report is explicit that its warning is not an argument against AI investment itself. The institution explicitly states that AI could raise productivity significantly over the coming decade and includes scenarios in its analysis where the technology's economic contribution genuinely justifies substantial continued investment. The distinction the report draws, and the one that should guide any responsible policy or investment response, is between believing in the underlying technology and believing in every financing structure that has been built around it. Canals, British railways, and the dot-com internet all involved genuinely transformative technology. Each was still overbuilt relative to what near-term returns could support, and the overbuilding, not the technology itself, produced the subsequent bust.

A responsible regulatory response, consistent with what BIS officials have recommended, centres on improving transparency rather than restricting investment. Regulators need substantially better visibility into the private credit and non-bank lending channels that have become central to AI financing, since these channels currently operate with materially less oversight than traditional bank lending despite now representing a comparable scale of systemic exposure. Disclosure requirements around circular financing arrangements — making the true extent of equity-debt-purchase commitment loops between hyperscalers, chipmakers, and AI labs visible to investors and regulators rather than obscured across multiple separate corporate filings — would directly address the opacity the BIS specifically flags as dangerous. Maintaining disciplined monetary policy, as the BIS recommends in its broader report, ensures that central banks retain the flexibility to respond to a credit shock without it compounding into the kind of inflation crisis that constrained policy responses during the 2008 crisis.

For investors and AI companies themselves, the responsible response is to treat the distinction between genuine productivity-driven revenue growth and revenue growth substantially dependent on continued capital expenditure by a small number of interconnected counterparties as the single most important variable in evaluating any individual company's risk profile. A company whose revenue depends heavily on continued purchasing commitments from one or two hyperscaler partners, financed through circular arrangements that could unwind simultaneously in a downturn, carries meaningfully more risk than a company with diversified, end-customer-driven revenue, even if both report similar headline growth figures today.

Conclusion

The Bank for International Settlements has not predicted that an AI bust will occur, and the institution is careful to frame its analysis in terms of risk and vulnerability rather than certainty. What it has done, with the institutional credibility that comes from a documented history of identifying systemic financial risks before they became visible to the broader market, is describe a specific and concrete mechanism by which a disappointment in AI returns could travel beyond technology stock valuations and into the credit system that underpins the broader global economy. The scale of the spending — more than $1 trillion from the five largest hyperscalers across two years — combined with the structural shift toward debt financing, the explosive growth of private credit lending to AI companies, and the opacity of circular financing arrangements, together describe a financial architecture that is meaningfully more fragile to an AI-specific shock than the equity-financed dot-com boom of two decades ago.

The appropriate response to that warning is neither panic nor dismissal. It is the kind of careful attention to financial plumbing that the BIS has spent decades advocating for, applied now to a sector whose technological promise is genuine but whose financing structures, by the BIS's own technical assessment, have grown faster and more interconnected than the transparency and oversight needed to manage them safely. The history the BIS report situates this moment within — canals, railways, the dot-com internet — offers a consistent lesson: the technology usually survives. The financing structures built around it, and the institutions that bet too heavily on those structures without understanding their interconnections, often do not.