On October 13, the International Monetary Fund (IMF) warned that if AI-related companies' earnings or investment returns fall short of market expectations, a sharp stock market correction could spread through the global financial system. In its Global Financial Stability Report, released in Bangkok, the IMF lists a correction in AI-related markets as a major risk, alongside a sharp rise in government bond yields and cyberattacks.
Huge investment in AI is propping up the global economy, but if that investment falters, the effects could be far-reaching. Fewer orders for data centers, weaker consumption as share prices fall, and asset sales by funds that have used borrowed money to expand their investments: if these happen in sequence, trouble in the AI industry could spread to other industries and countries.
The IMF is not, however, predicting an imminent financial crisis. In releasing the report, it also acknowledged that financial markets have withstood a number of shocks so far. The warning points to the danger that the investment and financing structures that have expanded on AI expectations could amplify losses if they fail to deliver the anticipated results.
AI investment keeps growing, but its profitability has yet to be tested
The IMF's World Economic Outlook, released on October 13, projects global growth of 3.1% in 2026 and 3.4% in 2027. Growth continues in the baseline scenario, and AI-related investment is one of its supports. But how much profit today's massive investment will generate has not yet been adequately tested.
As data center construction advances, demand for AI chips and power equipment grows, and sales rise at construction firms and equipment makers. Capital spending energizes economic activity even before AI services themselves turn a sufficient profit.
The IMF's World Economic Outlook also concludes that expanding AI-related investment and rising exports are supporting the current economy. Whether AI adoption will raise productivity across a wide range of industries, however, depends on results still to come.
In any case, it takes time to recover capital spending through profits. Companies install equipment first and must then recoup the costs through higher sales or more efficient operations. That requires businesses adopting AI to put the technology to real use and gain sufficient economic benefit.
The IMF's analysis published in April also notes that the economic impact of AI depends on more than the performance of cutting-edge models. How quickly and widely the technology spreads matters, as does whether institutions and infrastructure are in place. The arrival of a high-performing AI model does not by itself guarantee that investment across the economy will be recouped.
AI computing equipment also carries ongoing replacement costs.
In its January 19 analysis, the IMF warned that borrowed money makes up a growing share of AI investment, and that the depreciation periods assumed for high-performance processors greatly affect profitability.
If equipment has to be replaced sooner than expected, profits could be squeezed and additional borrowing may be needed. Even an AI service that is useful to its users does not necessarily generate enough profit to justify the investment in the equipment behind it.
Three channels through which an AI investment slowdown could spread to the global economy
If market expectations for AI investment collapse, the impact would not stop with AI companies and chipmakers.
Organizing the analysis in the market-correction scenario in Chapter 1 of the World Economic Outlook (pp. 28–30) and in Chapter 2 of the Global Financial Stability Report (pp. 23–25) reveals three main channels of transmission.
| Channel | First change | Impact on other industries and countries |
|---|---|---|
| Capital spending and trade | AI business profit outlooks worsen and capital spending shrinks | Orders for chips and related equipment fall, affecting production and economic growth in exporting countries |
| Asset prices and consumption | Falling share prices erode the value of household holdings | Consumption cools, and sales fall even at companies with no direct link to AI |
| Borrowing and asset sales | Assets held with borrowed money lose value, requiring extra collateral or repayment funds | Funds sell other assets too, causing market turmoil and spreading losses to financial institutions |
These are not independent risks; they could reinforce one another. Note, too, that the IMF is not simply adding up the losses through the three channels into a single forecast.
First, a cut in capital spending has effects across borders. If a major US technology company scales back its data center construction plans, for example, orders could also fall for Asian chipmakers and related equipment makers.
Second, falling share prices affect household consumption. If a sharp drop in AI-related shares reduces the value of holdings, households may cut spending, and the effects could extend to retailers and service businesses with no direct dealings with AI companies.
Third, in financial markets, investing with borrowed money could magnify losses.
When a fund that invests with borrowed money faces a decline in the value of its holdings, financial institutions and other lenders may demand additional collateral. This is known as a margin call.
If multiple funds hold large positions in the same stocks and are asked for more collateral at the same time, sales could concentrate and accelerate price declines.
Moreover, the assets sold to raise cash are not necessarily the AI-related shares that caused the losses. Other stocks and bonds that are easier to convert to cash may be sold instead, spreading turmoil to markets that have no direct connection to AI.
Chapter 2 of the Global Financial Stability Report analyzes the danger that leveraged investing, and cash-flow strains from investor redemption requests, could accelerate a chain of falling asset prices and sales.
Beyond borrowed money, derivatives can also allow investment on a scale beyond a fund's own capital. That magnifies losses when the market moves against expectations.
As a concrete example, the IMF cited a case in July 2026 involving the hedge fund Situational Awareness.
According to the report, the fund had concentrated investments in AI-related shares on a scale exceeding its own capital. As losses emerged it was asked for additional collateral and reduced its positions. Another market participant then took over its listed equity holdings, and the unwinding proceeded with turmoil contained.
This case did not lead to a global financial crisis. But it shows that the mechanism by which falling AI-related shares strain a fund's cash flow and force asset sales has already surfaced as a real-world problem.
The risk of "circular deals" that tie investment in AI firms to sales
The AI market also has features that set it apart from ordinary capital spending. In some cases, the company providing funds is also the one selling products or services to the company that receives them.
A study of circular relationships among AI firms, published October 1 by researchers at the Bank for International Settlements (BIS), analyzes this overlap of investment and commercial dealings.
For example, if a chipmaker or cloud provider invests in an AI model developer, the company receiving the funds can use them to buy the investor's chips or cloud services. In other words, the company providing the investment capital also earns sales from that capital.
A research team including Jon Frost examined 1,246 AI companies, matching 2021–2025 investment data against supplier and customer relationships between firms.
They found that among investment deals between AI firms, those that also involved a commercial relationship in the same period accounted for 16.1% by number of deals and 46.4% by the total value of disclosed deals. This suggests that a large amount of investment money is concentrated in a relatively small number of deals.
Care is needed in interpreting these figures, however.
The study's "circular relationships" broadly capture overlap between investment and commercial dealings over five years. They are not a tally of only those deals in which an investment contract includes an obligation to purchase products.
The investment amounts also use totals announced in funding rounds and similar events. For deals with multiple investors, each investor's individual contribution cannot be determined.
It would therefore be a mistake to read the 46.4% figure as meaning that 46.4% of AI demand was generated by money circulating within the industry.
There are also reasonable grounds for investment and commercial dealings to overlap.
Chipmakers and cloud providers can more easily see customers' usage and future demand, so they may be better placed than outside financial institutions to judge a business's growth prospects. For companies that buy chips and computing resources, investing in a supplier has the advantage of securing the equipment they need on a stable basis.
The BIS researchers do not treat such relationships as uniformly problematic.
But if a customer's business falters, the supplier could suffer a double loss: a fall in the value of the company it invested in, and a drop in future sales.
Moreover, even when sales across the market are growing, it becomes harder to tell whether that growth comes from end demand outside the AI industry or is propped up by investment among firms.
What the BIS research highlights is the possibility that such interdependence amplifies risk in the AI market.
If AI investment stalls, global GDP could come in 1.1% lower
The IMF's World Economic Outlook also estimates the economic impact if expectations for AI investment prove badly off the mark.
In that scenario, global GDP in 2028 would be 1.1% below the baseline scenario, and US GDP would be 1.8% lower.
The key point is that this figure is not a drop in the annual growth rate but the gap in the level of GDP compared with a case in which no shock occurs.
The estimate assumes that AI-driven productivity gains fail to materialize as expected and related investment falls sharply.
It also sets conditions in which US share prices fall 20% and those in other regions fall 15% in 2026. It further factors in rising credit-risk spreads on corporate borrowing rates and unease spreading to government bond markets.
In short, this is a simulation of what could happen if several adverse conditions coincide. It is not an IMF forecast of a 20% stock decline in 2026 or of the timing of any AI bubble burst.
The same analysis puts the direct impact of reduced AI-related investment on global GDP at a shortfall of up to 0.4% by 2028.
The 1.1% shortfall in overall global GDP, on the other hand, also includes the effects of reduced consumption from falling share prices and a worsening financial environment.
This suggests that the economic blow could be magnified less by the cut in AI equipment investment itself than by its spillover into consumption and financial markets.
The IMF is not, however, considering only pessimistic scenarios. It also analyzes a scenario in which AI's productivity gains arrive sooner than expected and boost economic growth.
The report does not dismiss the economic value of AI technology. It examines how far the financial system could absorb the impact if market expectations for AI growth were to change.
Can banks withstand a crisis? Policy responses have limits
The IMF has also tested the resilience of the financial system in a case where serious shocks hit bond and energy markets at the same time as turmoil in AI-related markets.
It concluded that the banking sector as a whole could withstand even severe stress tests. Some weaker banks, however, could face capital shortfalls. Even if the financial system as a whole can survive a crisis, that does not mean every bank is safe.
Financial institutions other than banks also cannot be ignored when considering the impact on financial markets.
According to the IMF, total assets in the hedge fund industry reached about $13 trillion in early 2026. That is the industry's total assets, however, and does not represent the amount invested in AI-related assets.
The IMF also pointed to growing investment by insurers and pension funds in assets that are hard to convert to cash, and rising risks tied to borrowing and liquidity in the private credit market, where non-bank investors lend directly to companies.
These institutions and funds support the supply of capital to companies in normal times. But if market conditions worsen, asset sales and reduced lending could end up deepening the damage to the economy.
Whether governments and central banks can then provide sufficient support is also a crucial question.
The IMF noted that rising public debt and inflation in many countries have narrowed the room for policy responses if a financial crisis occurs, compared with the past.
If heavy government bond issuance coincides with private companies trying to raise huge sums, competition for limited investment funds intensifies. Even if markets are thrown into turmoil, interest rate cuts or public support will not necessarily bring them under control.
The threats to the financial system are also not limited to an AI investment slowdown.
The IMF's June analysis of AI and cybersecurity in the financial sector notes that when multiple financial institutions depend on common IT infrastructure or external service providers, failures and cyberattack damage can spread in a chain reaction.
Even if banks hold sufficient capital, financial services can be disrupted if a cyberattack halts payment and trading systems. Financial soundness and the ability to keep systems running reliably each need to be tested.
What to watch to judge whether AI investment is sustainable
In light of the IMF's warning, assessing the AI market means looking not only at model performance and user growth but also at whether the money invested can be recovered.
Particularly important are whether sales from customers outside the AI industry are growing sufficiently, whether companies can secure profits and cash flow while continuing capital spending, and whether chip replacement cycles and debt repayment deadlines are manageable.
Contracts such as the "residual value guarantees" discussed in the BIS research also deserve attention. These are arrangements that cover the shortfall if the value of equipment after a certain period falls below expectations. The existence of such contracts changes who ultimately bears the loss when equipment values drop.
If AI companies secure sufficient profits and can use them to buy and replace equipment and repay debt, today's massive investment could become a foundation for future economic growth.
If, on the other hand, sales growth depends heavily on funding supplied among AI firms and recovery of capital spending does not progress, the effects of a pullback in expectations may not stop at the technology industry.
In judging whether the huge investment in AI is sustainable, what will matter going forward is not only the flow of money within the industry but real demand from a wide range of industries and the earnings growth that demand supports.