AI promises a more productive world, but we must manage risks in the transition.
AI is here, and already having a profound impact on the global economy.
Investment is flowing into AI on a scale that will likely exceed the vast sums that went into building the railroads, the electricity grid, and the internet. In the short run, the AI boom is a positive demand shock that is lifting growth, especially in the U.S. and Asian countries that supply hardware. Yet not all countries are riding this wave, and without broader adoption, too many could be left behind.
While the demand shock is happening now, the real promise lies in the future: the prospect of a positive supply shock that could significantly boost productivity and raise global growth. How much is hard to predict, but the growth boost could be on the order of half a percentage point annually over the next decade. It may take time for the productivity gains to materialize—technological transformations rarely happen overnight. To unlock the benefits of AI, businesses must rethink, reinvent and re-engineer their product lines, processes, and workflows. Employees and consumers must find the best ways to use new models. The benefits of the internet did not come with the ability to throw up a home page. The real gains came with the emergence of new business models and the incremental improvements that connectivity enabled.
This time, much of the boom is being driven by massive investment in AI infrastructure such as data centers, with hyperscalers leading the race. While these large companies have profitable core businesses, they are increasingly turning to debt markets for financing. Just in the last two years, their share of investment-grade bond issuance has increased more than tenfold—from less than 1 percent to 10 percent and growing. This is happening in an industry that is highly concentrated and where dealmaking is often circular, with hyperscalers investing in frontier-model labs, which in turn buy data capacity from those same firms.
The risk is that expectations may run rapidly ahead of productivity gains, setting the stage for near-term disappointment. According to Amara’s Law, we tend to overestimate the effect of a new technology in the short run and underestimate its effect in the long run. Add to this concerns about increasingly powerful models escaping human control or undercutting cybersecurity. At some point, risks may converge to challenge macroeconomic and financial stability.
What might this look like? Big revenue or earnings disappointments by AI developers could undermine expectations of future profitability, dealing a hit to equity prices. Unsettling incidents involving frontier models or cyberattacks could dent confidence and raise financial stability risks. In a severe scenario, we could see a market correction that cascades through the highly leveraged and circular AI ecosystem. Given the sizable cross-border holdings of AI assets, the losses could spread around the world.
The disruption to the job market could also be profound. According to our research, roughly 40 percent of jobs globally are exposed to AI. In advanced economies, exposure rises to about 60 percent. Within these bands of exposure, some jobs will likely be augmented, while others may be replaced by the technology. It is still too early to predict the toll on employment—so far, there is no apparent impact on the job market, but the nature of work is changing, and this is likely to accelerate.
And then there are the cybersecurity risks. As finance becomes more digital and interconnected, cyber disruptions are emerging as growing source of systemic risk. AI adds urgency by accelerating vulnerability discovery, exploitation, and attack, compressing the attack-defense cycle to machine speed. A major attack affecting payments, trading, or access to deposits could cause immediate losses and undermine confidence. If an operational disruption strikes a critical node, including outside the financial sector—for instance, affecting critical infrastructure—it could quickly evolve into a broader confidence shock.
We must be ready to face and mitigate these risks. We also must avoid a scenario where the gains from AI are reaped only by a privileged few. This applies to individuals and countries alike. Indeed, the growth increase in advanced economies may be more than double the increase in low-income countries, reflecting gaps in access, preparedness, and adoption.
So, in practice, what can policymakers do?
Countries need policies that are grounded in the unique characteristics of AI, but agile enough to account for its uncertain path and rapid pace of change. Here are three core principles that can light the way.
First: safeguard macroeconomic and financial stability.
As with any shock, the priority should be to preserve stability.
The good news is that regulators and supervisors don’t need to start from scratch. Existing tools, including capital and concentration requirements, can contain the buildup of risks in the financial system. Generally speaking, the goal should be to encourage broader participation in the financing of the AI ecosystem. Countries should bolster defenses against cyber risks, such as through resilience testing of banks. Regulators have stepped up international coordination on these issues—that should continue.
Central banks must stay on their toes, as the impact of AI on inflation will vary from country to country. Maintaining price stability is essential, especially at a time when other forces are adding inflation pressure.
On the fiscal side, governments should reexamine their medium-term plans through the lens of the AI transition. AI may raise growth, but it could also reduce labor income, increase inequality, and erode the tax base. Countries will need fiscal space to invest in infrastructure and skills, and cushion the impact on the most vulnerable. And they will need tax systems designed for the AI era. With debt already high, policymakers will need to make hard choices.
Finally, governments and the private sector must rapidly scale up their cyber defenses. Addressing cybersecurity risks requires a systemwide operational resilience framework focused on the continuity of critical services. Policy priorities should be tailored to the levels of cyber maturity of different economies.
Second: enable adoption and innovation.
Reliable digital infrastructure and a skilled workforce will be the foundation for countries to successfully adopt AI. Here, the global gaps are large. Advanced economies are three times better equipped in digital infrastructure than low-income countries, according to our AI Preparedness Index.
Many economies will also need more energy infrastructure to support AI deployment, which will sharply raise electricity demand. By 2030, the International Energy Agency expects data centers to account for 3 percent of global electricity consumption. To put that in perspective, demand from data centers by then will exceed Japan’s power consumption today. In the AI age, countries will need a medium-term strategy to reliably increase power supply.
But it will take more than better infrastructure and a skilled workforce to fully unlock the productivity gains from AI. Many economies need a deeper well of private risk capital to finance startups and encourage innovation. In countries where governance and state capacity are strong, private capital could be complemented by public-sector investments in AI research and development. In Sub-Saharan Africa, for example, mobilizing private finance for AI infrastructure requires both strengthening the business climate and providing targeted government subsidies and guarantees.
A dynamic AI economy should allow the smooth entry and exit of firms. Companies must be able to reorganize and reallocate resources so AI measurably improves their business. Too often, they face structural tax distortions that favor legacy investments, and red tape that smothers growth and squeezes profits. Governments can help by modernizing their tax systems, reforming their insolvency regimes, and improving legal certainty.
Third: support workers and create opportunities.
AI will take over some tasks done by workers. That doesn’t necessarily mean those workers will lose their job. They may move into a new role with a different bundle of responsibilities that can be augmented by AI. They may even get a better job with a different firm, as the gains from higher productivity spread throughout the economy.
IMF research on Asia-Pacific highlights some of these forces. Demand from employers in the region has declined about 10 percent in highly AI-exposed occupations. At the same time, demand is growing for workers who can use AI to do more. New roles are emerging, such as prompt engineering and building AI agents, that didn’t exist a few years ago.
Labor markets should be flexible enough that firms can recruit the talent they need. At the same time, workers displaced by the technology may require support. The policy challenge is to find the right balance between these two needs. We must equip people, especially our young, with cognitive, creative, and technical skills that empower them to use AI. Traditional social safety nets such as unemployment insurance will remain important. But policymakers will need to increasingly deploy active labor-market programs to upskill and reskill workers. Governments should calibrate their tax systems so they do not excessively favor or impede automation.
These principles should help countries navigate the AI age. Of course, countries cannot and should not go it alone. Coordination among governments, AI companies and other stakeholders will be critical, given the rapidly growing capabilities of frontier models and the interconnectedness of countries. At the IMF, we will continue to offer policy advice, capacity building and, when necessary, financial support within our core macroeconomic mandate.
The world is going through a technological transition that may force us to reconsider how we protect critical infrastructure, how we preserve financial stability, and even basic ideas such as how we work. That can be scary. We must take public concerns about AI seriously—people understandably want to know how it will affect them and their families. But AI also holds the promise of a more productive world where humans, aided by technology we invented, achieve things we never would have thought possible.
Let us embrace the challenge together, and learn as we go.