A broader reckoning for the A.I. boom

The artificial intelligence boom is entering a more sober phase.

After months dominated by eye-popping valuations, blockbuster chip demand and corporate jockeying, a different set of questions is now moving to the foreground: How much of the industry’s financing is built on fragile assumptions? What happens to the entry-level jobs that have long fed finance and consulting? And how much climate damage is being locked in by the race to build the computing infrastructure behind the technology?

Those concerns converged on Sunday as the International Monetary Fund warned that “circular financing” inside the A.I. ecosystem had swollen to just over $1 trillion, while new reporting and research pointed to upheaval in white-collar labor markets and a steep environmental cost from the expansion of data centers.

The shift in attention reflects a larger change in how investors, regulators and executives are beginning to judge A.I.: less by promise alone, and more by whether the economics, labor effects and environmental trade-offs can withstand scrutiny.

IMF warns of interlocking risks

Kristalina Georgieva, the managing director of the I.M.F., said the global economy was already facing unusual strains and cautioned that financing patterns inside A.I. were becoming a source of vulnerability in their own right.

Her warning centered on what the fund has described in recent months as an increasingly concentrated A.I. system — one dominated by hyperscale cloud providers, model developers and chip suppliers whose fortunes are bound together through cross-investments, borrowing and long-term revenue commitments. The result, the I.M.F. has argued, is a web of dependencies that can magnify shocks if demand falters or valuations fall.

According to Georgieva, circular financing in the sector has now climbed above $1 trillion, with roughly two-thirds of it debt-funded. In practical terms, that means A.I. companies are not simply attracting outside capital; they are also increasingly financing one another’s growth through arrangements that can make balance sheets across the sector more interdependent.

Such structures are not inherently unsustainable. But they leave the industry more exposed if one of its key assumptions weakens — whether on customer demand, pricing power or the timetable for turning costly A.I. investments into durable profits.

That question has become more pressing as enthusiasm cools from its most feverish highs. While chipmakers and infrastructure providers have been clear beneficiaries of the spending surge, the broader business case for A.I. remains less settled, especially for companies still searching for revenue models that justify the cost of training and running advanced systems.

White-collar work begins to change

Nowhere is that uncertainty more visible than in the workplace.

European banks are increasingly deploying A.I. to automate routine administrative and analytical tasks, according to reporting on the sector, allowing them to trim some functions, redeploy employees and increase hiring for more technical roles. The technology is being applied first where repetitive work is abundant: document processing, compliance support, customer-service tasks and portions of internal analysis.

That pattern is consistent with a wider trend across professional services, where companies have found it easier to use A.I. to absorb junior-level tasks than to replace senior judgment. In banking, that means the technology is less a wholesale substitute for experienced dealmakers or risk managers than a tool that alters the structure beneath them.

Consulting firms are confronting a similar disruption. For decades, the industry’s business model has rested on armies of junior staff performing research, drafting presentations and handling other labor-intensive work that could be billed to clients while training the next generation of partners. A.I. threatens to compress both parts of that model at once.

If software can complete more of those early-career assignments in minutes, firms may face pressure on billable hours as well as on the traditional graduate pipeline that has long supplied them with talent. That has raised uncomfortable questions inside the industry about whether companies will keep hiring young workers at the same scale if fewer entry-level tasks remain to be done.

The answer is not yet clear. Some executives argue that A.I. will free junior staff from rote work and allow them to develop higher-value skills faster. Others fear the opposite: that the first rung of the ladder may simply narrow.

Either way, the labor implications are beginning to look less theoretical. Rather than producing an immediate wave of mass layoffs, A.I. appears to be redrawing career paths in sectors that once offered predictable apprenticeships into the professional middle and upper classes.

The environmental cost of scale

At the same time, the physical infrastructure required to sustain the A.I. expansion is drawing sharper criticism.

A Harvard-led study published in *Nature Sustainability* found that 403 U.S. hyperscale data centers emitted about 25.7 million metric tons of carbon dioxide from May 2024 through April 2025 — roughly equivalent to Denmark’s annual emissions. The researchers estimated that those facilities consumed about 82 terawatt-hours of electricity during the period, with more than half of the power coming from fossil fuels.

The findings underscore a basic tension at the center of the A.I. buildout. Technology companies have portrayed artificial intelligence as a tool that could accelerate scientific research, improve efficiency and even aid climate solutions. But the enormous computing load required to train and deploy advanced models is driving a parallel surge in electricity demand, much of it met by grids that remain heavily dependent on gas and coal.

The issue is especially acute for hyperscalers, the giant facilities that underpin cloud computing and A.I. systems. Their scale allows them to deliver vast processing capacity, but it also means new construction can quickly overwhelm local clean-energy supplies, forcing operators to rely on dirtier power sources to keep pace with demand.

The study’s authors focused on a period ending in April 2025, before some of the latest wave of expansion. That leaves open the possibility that emissions could climb further unless efficiency gains, cleaner power procurement and better siting decisions begin to outpace growth.

From hype to accountability

Taken together, the developments suggest that the next phase of the A.I. era may be defined less by spectacle than by accountability.

The financing warning from the I.M.F. points to a potential financial-stability problem if interlocking bets across the industry start to unwind. The changes in banking and consulting hint at a quieter but potentially profound labor-market shift, one that could alter how elite white-collar professions recruit and train young workers. And the emissions data indicate that the race for computational power is colliding with climate goals in ways that are becoming harder to ignore.

None of this means the A.I. boom is ending. Companies are still spending heavily, and many executives remain convinced that the technology will transform productivity over time. But the burden of proof is changing.

For a while, the central question was whether artificial intelligence was powerful enough to justify the excitement. Increasingly, the harder question is whether the systems being built around it — financial, corporate and physical — are resilient enough to justify the cost.

Sources

Further reading and reporting used to add context: