The Bank for International Settlements has issued a warning that will feel familiar to anyone who has watched previous technology cycles run hot and then cool sharply: when expectations outrun measurable results, funding can tighten faster than companies can adjust. In a new assessment, the BIS points to a risk that “AI exuberance” could end not with a gentle normalization, but with a more abrupt investment bust—one that would not stay confined to data centres and chipmakers, but could ripple through credit markets, labour demand, and broader economic activity.
The core of the BIS concern is not that AI will fail. It is that the market’s pace and scale of enthusiasm may be outstripping the pace at which returns become visible, comparable, and durable. When investors believe the future is arriving immediately, they tend to fund growth aggressively—often on the assumption that profitability will follow later. But if the returns are weak, uneven, or delayed, capital can retreat quickly. That retreat matters because AI spending is not just a line item; it is increasingly tied to financing structures, supply chains, and balance sheets across the financial system.
What makes this warning distinctive is the BIS framing. Rather than treating AI as a purely technological story, it treats it as a financial stability issue—an area where timing, valuation discipline, and liquidity conditions can matter as much as innovation itself. The BIS is essentially asking a question that investors sometimes avoid until it becomes unavoidable: what happens if the “investment thesis” for AI does not translate into cash flows quickly enough to justify the cost of capital?
To understand why the BIS is sounding the alarm now, it helps to look at how AI investment differs from earlier waves of technology spending. Many past cycles were driven by hardware upgrades, enterprise software rollouts, or consumer adoption. AI, by contrast, often requires large upfront expenditures—compute, energy, specialised chips, cloud capacity, and talent—before the benefits are fully realised. Even when models improve rapidly, the business value depends on integration: workflow redesign, data governance, compliance, and the ability to convert model outputs into reliable products and services. That conversion can take time, and it is rarely uniform across industries.
The BIS warning therefore targets a mismatch between two timelines. One timeline is the market’s expectation of near-term payoff. The other is the operational timeline of turning AI into measurable productivity gains, revenue growth, or cost reductions. If those timelines diverge too far, the funding environment can change abruptly.
A key mechanism behind the BIS risk is the speed at which capital can tighten. In boom periods, investors often treat AI exposure as a momentum trade: valuations rise, funding becomes easier, and the cost of capital falls relative to expected growth. But when performance disappoints—even modestly—investors can reprice risk quickly. That repricing can show up first in secondary markets, then in primary funding rounds, and finally in the real economy through hiring freezes, delayed capex, and renegotiated supplier contracts.
The BIS highlights that returns may be “weak or uneven.” Uneven is important. Even if some firms demonstrate strong ROI, others may struggle due to data limitations, integration challenges, or simply because their use cases do not scale economically. In such a scenario, investors may not withdraw uniformly; they may instead concentrate funding toward the perceived winners and starve the rest. That selective tightening can still produce systemic effects, because the AI ecosystem is interconnected. Suppliers depend on demand from major buyers; smaller firms depend on venture funding; and lenders depend on the stability of borrowers’ cash flows.
This is where the BIS’s “investment bust” language becomes more than a metaphor. An investment bust is not merely slower growth. It is a contraction in the flow of funds that can force companies to cut spending abruptly, even if they still believe in the long-term potential of AI. When funding dries up, the option to “wait for the next quarter” disappears. Companies must preserve liquidity. They reduce burn rates. They renegotiate terms. They delay projects. And they may also face refinancing risk if their capital structure was built on optimistic assumptions.
The global spillover risk comes from the fact that AI investment is increasingly financed through channels that extend beyond venture capital. Public markets, corporate credit, private credit, and bank lending all play roles in funding technology expansion. If valuations compress and risk appetite declines, the cost of capital rises across the board. That can affect not only AI firms but also adjacent sectors—semiconductors, cloud infrastructure, cybersecurity, data management, and professional services—that rely on sustained tech spending.
There is also a second-order effect that the BIS implicitly points toward: the feedback loop between funding conditions and expectations. When investors see early signs of weaker returns, they may lower their forecasts for the entire sector. Lower forecasts can lead to lower valuations. Lower valuations can make it harder to raise new capital. Harder fundraising can lead to reduced spending. Reduced spending can then lead to fewer measurable improvements, reinforcing the initial disappointment. This is how a “bust” can become self-reinforcing.
Yet it would be misleading to frame the BIS warning as a prediction that AI will stop delivering value. The BIS is not arguing against AI’s utility. It is warning about the financial consequences of overconfidence. In many technology transitions, the first wave of deployments produces mixed results. Some firms capture outsized gains; others learn through costly experimentation. Over time, best practices emerge and ROI becomes clearer. The danger is that the financial system may not be patient enough to allow that learning curve to play out.
One unique angle in the BIS approach is its attention to the possibility of a “sharp pullback.” Sharpness matters because gradual cooling allows companies to adapt. A sudden pullback forces adaptation under stress. It can also create liquidity problems. Even profitable firms can face cash flow timing issues if capital markets freeze. And even promising firms can be forced to accept unfavourable terms or dilute heavily to survive.
The BIS warning also resonates with a broader pattern seen in recent years: the tendency for markets to price long-duration assets with short-duration impatience. AI investments often resemble long-duration bets. They require sustained funding and patience for returns. But when macroeconomic conditions tighten—through higher interest rates, tighter credit standards, or risk-off sentiment—the market’s willingness to fund long-duration bets can shrink quickly. In that environment, AI exuberance can become fragile.
So what should investors, policymakers, and company leaders watch next? The BIS’s warning suggests several practical indicators.
First, evidence of real, sustained ROI from AI deployments. Not “pilot success,” not impressive demos, and not isolated case studies. The market will look for repeatable metrics: measurable productivity improvements, reduced operating costs, increased conversion rates, improved customer retention, or new revenue streams that persist across quarters. Importantly, ROI needs to be evaluated net of total costs, including compute, data preparation, integration, and ongoing model maintenance. Many early deployments underestimate these costs, especially when scaling from a controlled environment to production.
Second, investor signals on funding and valuations. If the sector is healthy, funding should remain available even as valuations normalise. If the BIS risk is materialising, investors may begin to demand stronger proof earlier, reduce round sizes, and shift from growth-at-any-price to survival-at-all-costs. You may also see changes in deal structure: more preference for secured terms, more emphasis on milestones, and more pressure on burn-rate discipline.
Third, how quickly companies adjust spending if demand or returns soften. This is a behavioural indicator. Firms that can scale down without damaging their core capabilities are less likely to trigger cascading failures. Firms that cannot—because they have committed to long-term compute contracts, expensive capex, or staffing plans—are more exposed to funding shocks. The BIS warning implies that the speed and flexibility of adjustment could determine whether disappointment remains contained or becomes a broader contraction.
There is another dimension worth considering: the distribution of returns across the ecosystem. If AI value accrues mainly to a small number of platforms or model providers, then downstream firms may struggle to monetise their own deployments. That would create a structural imbalance. Downstream companies might keep spending to catch up, but without sufficient revenue, they become dependent on continued external funding. In a funding bust scenario, that dependence becomes a vulnerability.
Conversely, if value creation is more distributed—if enterprises can capture productivity gains and share them internally—then the sector may be more resilient. The BIS warning therefore indirectly raises a question about business models: are AI investments translating into durable economic value for the firms making the investments, or are they primarily enriching upstream suppliers and platform owners?
The BIS also implicitly invites a closer look at risk management inside companies. In exuberant periods, firms can underestimate the volatility of demand and the uncertainty of ROI. They may hire aggressively, lock in compute capacity, and build product roadmaps based on optimistic assumptions. When returns disappoint, the problem is not only that profits are lower; it is that the cost base is already set. That is why liquidity planning and scenario analysis matter. Companies that treat AI spending as a portfolio—balancing experiments with disciplined scaling—may weather downturns better than those that treat it as a single, linear path to profitability.
For policymakers, the BIS warning suggests that AI should be monitored like other sectors where financial conditions can amplify economic swings. That does not mean regulating AI itself in the traditional sense. It means paying attention to credit conditions, leverage, and concentration risks. If AI-related exposures become concentrated in certain parts of the financial system—such as private credit funds, leveraged balance sheets, or highly valued but cash-constrained firms—then a repricing could have outsized effects.
There is also a communications challenge. When central banks and regulators talk about AI, they can inadvertently contribute to hype. The BIS warning, by contrast, is a reminder that financial systems respond to narratives, and narratives can become dangerous when they are not anchored to measurable outcomes. The BIS is effectively urging a shift from excitement to verification.
Still, it is worth acknowledging the upside embedded in the BIS caution. A funding bust is not inevitable. In many