The stock-market frenzy around artificial intelligence may be losing steam, but a less visible and potentially more dangerous expansion could be building underneath it. Analysts at the Brookings Institution and officials at the International Monetary Fund have drawn a distinction between the AI hype cycle that lifted equity valuations and a quieter buildup of risk in physical infrastructure, private lending, and energy commitments tied to AI. The difference matters because the second kind of bubble is harder to spot and harder to unwind.
Two Bubbles, Not One
Most public debate about an AI bubble has focused on stock prices and whether companies like Nvidia or Microsoft can justify their valuations with actual revenue. That conversation, while valid, misses a structural shift happening off public exchanges. A recent Brookings analysis lays out six concrete indicators for testing whether an AI bubble exists: investment volumes, data-center construction timelines, enterprise adoption rates, pricing trends, competitive dynamics, and user trust. Using those indicators, the Brookings piece argues that some signs of froth can cool as investors grow more selective about which AI bets will pay off.
Yet the same analysis flags a rarer type of bubble that may still be inflating. This one lives in infrastructure build-out, private credit exposure, and adoption metrics that have not yet been tested by a downturn. Data centers require years of planning and billions in capital commitments that cannot be easily reversed. Private lenders, not regulated banks, are financing a growing share of these projects. If AI adoption stalls or energy costs spike, the losses will not show up on a stock ticker first. They will surface in loan defaults, stranded assets, and regional power-grid strain.
IMF Leadership Flags Concentrated Optimism
IMF Managing Director Kristalina Georgieva has been pressing a related but distinct point: the optimism driving AI investment is dangerously concentrated. In a speech titled “Opportunity in a Time of Change,” delivered as a curtain-raiser for the IMF’s 2025 Annual Meetings, Georgieva framed AI as holding real promise while warning that promise alone is not enough. The technology must translate into measurable productivity gains and broader economic growth, she argued, or the investment wave risks becoming a drag rather than a driver.
That theme sharpened during a press briefing on the Global Policy Agenda, where Georgieva addressed the “AI investment boom” directly. According to the official transcript published on October 16, she pointed to the concentration of optimism as a specific vulnerability. When capital flows into a narrow set of bets based on expectations that have not been validated by real-world output, she warned, the correction can be abrupt. In the briefing, she also emphasized that AI’s economic benefits depend on how widely the technology is adopted and whether workers and institutions can adapt fast enough to capture those gains.
Her framing carries weight because the IMF is not simply commenting from the sidelines. The Fund’s own research agenda is actively studying how AI investment interacts with financial stability, and the 2025 meetings have placed the topic at the center of multilateral economic discussions. The institution’s communications hub at the IMF press center underscores how frequently AI now appears in speeches, reports, and policy notes aimed at finance ministries and central banks.
Central Banks Sound Similar Alarms
The IMF is not alone in raising these concerns. The Bank of England also issued warnings about AI-related financial risk around the same time, creating an unusual moment of alignment between two influential economic institutions. As Associated Press reporting makes clear, both the BoE and the IMF flagged bubble dynamics in AI, though their specific emphases differed. The BoE focused more on financial-system exposure, while the IMF stressed the gap between investment enthusiasm and proven productivity returns.
When a central bank and a multilateral lender converge on similar risk concerns in close succession, it suggests the debate is moving beyond purely academic circles. These institutions typically choose their public warnings carefully. For investors, lenders, and policymakers, the practical takeaway is that scrutiny of AI-linked financial exposures could increase, especially where risks are concentrated or opaque.
Why the Infrastructure Bubble Is Harder to Pop
The distinction between a stock-market AI bubble and an infrastructure AI bubble is not just semantic. Equity bubbles, while painful, tend to correct relatively quickly. Prices fall, portfolios shrink, and capital reallocates. Infrastructure bubbles operate on a different timeline. A data center that takes three years to build and finance cannot be abandoned midway without crystallizing massive losses. Power-purchase agreements signed to feed those facilities lock utilities and developers into long-term obligations regardless of whether AI demand materializes at the projected scale.
Private credit adds another layer of complexity. Unlike publicly traded bonds, private loans to AI infrastructure projects are not marked to market daily. Their valuations depend on appraisals and assumptions that can lag reality by months or even years. If the Brookings indicators are right that froth in the most visible parts of the AI trade is cooling, the slower-moving exposure in private lending and physical assets may not register the same correction for some time. That delay creates a window where risk accumulates silently.
This dynamic echoes earlier technology busts, but with a twist. The dot-com crash of 2000 destroyed paper wealth. An AI infrastructure correction would hit physical supply chains, energy markets, and regional economies that depend on construction and operations jobs tied to data-center campuses. The real-world footprint is larger, and the exit ramps are fewer. Once a region has rezoned land, upgraded transmission lines, and built water and cooling systems around a cluster of facilities, reversing course is politically and economically costly.
What Productivity Has to Prove
Georgieva’s insistence that AI must deliver productivity gains is not a generic talking point. It is the specific condition that separates a justified investment wave from a misallocation of capital. If AI tools meaningfully raise output per worker, lower the cost of key services, and enable new business models, then today’s heavy spending on chips, data centers, and software could pay off in higher growth and tax revenues. In that scenario, even aggressive leverage might be manageable.
If those gains fail to materialize at scale, however, the same investments become a drag. Companies that borrowed heavily to “modernize” will be left with higher interest costs but no corresponding revenue lift. Governments that offered subsidies and tax breaks to attract AI infrastructure may find they have traded fiscal space for facilities that operate below capacity. The risk is not that AI produces no value at all, but that its incremental benefits fall short of the heroic assumptions built into current project pipelines and loan books.
Productivity, in other words, is the bridge between technological promise and macroeconomic reality. Without clear, broad-based improvements in how economies produce goods and services, the AI boom looks less like the early stages of a transformative general-purpose technology and more like a capital-intensive experiment that overshot.
Policy Choices That Could Deflate the Risk
None of this means that an AI infrastructure bust is inevitable. Policymakers and market participants still have room to steer the boom toward more sustainable ground. Regulators can push for better disclosure of AI-related exposures in both public markets and private credit, making it harder for risks to hide in opaque structures. Supervisors can stress-test banks and large nonbank lenders against scenarios where AI demand underperforms, energy prices spike, or both.
Governments, for their part, can design incentives that favor flexible, modular infrastructure over single-purpose megaprojects. Data centers that can be repurposed for broader cloud services, or that are sited in regions with diversified industrial bases, will be less likely to become stranded assets. Energy policy will matter as well: aligning AI-driven power demand with investments in resilient grids and low-carbon generation can reduce the chance that overbuilt capacity collides with climate or regulatory constraints.
For investors, the emerging message from Brookings, the IMF, and the Bank of England is not to flee AI altogether, but to distinguish more carefully between software bets that can scale up or down quickly and hard-asset bets that lock in multi-decade obligations. The first bubble may already be deflating in public markets. Whether the second, slower-moving bubble in infrastructure can be managed without a crash will depend on how quickly the world’s enthusiasm for AI is matched by equally rigorous attention to where, and how, the money is actually being spent.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
