From AI boom to economic bust
Ian Goldin
OXFORD — The divide between artificial intelligence (AI) fantasists and doomsters is growing. The fantasists believe AI will turbocharge growth and create a utopian world of abundance. The doomsters believe it poses an existential threat to humanity and that, before killing us, it will lead to mass unemployment and a world of pervasive surveillance and alternative realities. Who is right? And can governments do anything to mitigate the risk of catastrophe?
AI enthusiasts gush about the technology’s implications for productivity. Anthropic CEO Dario Amodei has predicted that U.S. GDP growth will rise from today’s paltry 2 percent to 10–15 percent. Elon Musk expects it to double U.S. growth to more than 4 percent next year and believes machines will soon replace humans in almost every occupation.
This is an illusion, if not a delusion. The data-center boom and soaring stock-market valuations may boost near-term growth, but their contribution will be short-lived. After all, much of the economic activity in developed countries takes place in sectors where AI will have little effect on productivity, such as child and elderly care, construction, hospitality, and personalized services.
True, AI could boost output in manufacturing and professions like law and banking, which together account for roughly 20 percent of the economy in both the United States and the United Kingdom. But even if AI doubles productivity in these sectors, total economic output would rise by only 20 percent.
How AI may hit growth
AI itself could drag down economic growth in several ways. For starters, data centers’ growing electricity demands could push up energy costs, keeping inflation and therefore interest rates higher than they otherwise would be.
Second, some of AI’s earliest benefits are likely to come from longer life expectancy, thanks to better management of chronic diseases like diabetes and faster drug discovery. This is good news, but it could also accelerate population aging and strain pension and care systems, as progress in treating dementia and neurodegenerative diseases lags far behind advances in other areas of medicine. A growing share of the workforce and public spending will be devoted to elderly care, at the expense of more productive sectors.
And then there’s the gargantuan $2.4 trillion being spent (with more on the way) on the AI buildout, fueled by ballooning debt and inflated private-equity and stock-market valuations. In the state of Ohio alone, Nvidia has committed $105 billion to the expansion of AI infrastructure. When, not if, the bubble bursts, the resulting market crash will invariably hit retirement savings and consumer confidence, slowing growth and potentially triggering a recession.
Moreover, the faster AI advances, the faster we will need to modernize our power grids, business practices, education systems, and much else. Yet investment rates in the United Kingdom and the United States are stagnant or declining. As more businesses and workers struggle to keep up, productivity growth will continue to slow. In many emerging markets, investment rates are almost double those of the U.K. and the U.S. These economies are also less burdened by outdated infrastructure and the costs of caring for aging populations, giving them a competitive advantage.
More losers than winners
AI will compound the political costs of slow growth. When economies stagnate, governments can meet urgent demands, such as higher defense spending or climate adaptation, only by cutting social security or other public services. With rising debt leaving little room for redistribution, and tax hikes deeply unpopular, governments will increasingly be unable to compensate those who lose their jobs to AI.
Far from AI creating a more level playing field, AI is concentrating income and economic activity. In an AI-driven knowledge economy, only a limited number of places can become innovation hubs. Dynamic firms benefit from proximity to one another and to cultural amenities that cater to highly skilled workers, concentrating the benefits of AI in fewer places and widening geographic disparities. Since remote work undermines spontaneous creativity and limits informal learning and the development of professional networks, it is unlikely to spread jobs and opportunities widely enough to reverse the decline of left-behind communities.
Governments should guarantee conditions for a decent life everywhere, but no industrial policy can create jobs everywhere. Trying to “level up” struggling regions by taking resources away from dynamic ones risks leveling down the entire economy. A far better approach is to make it easier for people to move to where the jobs are through more affordable housing and transport. People in the US, UK, and Northern Europe today are half as mobile as they were 50 years ago, largely because of high childcare and housing costs and unreliable public transportation.
The prospect of an AI “jobocalypse” is giving people good reason to be anxious. My own research group at Oxford estimates that up to 50 percent of all jobs in the UK and the US are vulnerable to automation. With new jobs requiring different skills and largely located far from those being destroyed, it is hardly surprising that most workers see AI as a threat to their way of life.
The rush to build data centers is also fueling a mushrooming populist revolt against AI. Besides being blamed for rising energy prices, they are noisy, consume vast amounts of water, and offer few long-term jobs to local residents. As more communities face economic decline, more elected officials will have to confront furious constituents demanding that something be done.
In addition to exacerbating inequality and fueling extremist politics, AI is inflaming geopolitical tensions. The belief that AI development is a winner-takes-all race is quite possibly the only idea that retains bipartisan support in Washington, with fears of Chinese AI supremacy being used to justify deregulation and give Silicon Valley free rein to “move fast and break things.”
Agents of destruction
The idea that any one country can win this race is misguided. No matter how much a country invests or how large its pool of skilled workers, advances will quickly be copied and improved upon elsewhere, and no government can manage the risks posed by AI on its own.
An AI-engineered pandemic, for example, would not respect national borders. Nor would an AI agent that escapes human control be any less dangerous if it came from the US instead of China. The only way to contain these threats is through international cooperation. US restrictions on chip exports to China and scientific collaboration, however, are making cooperation between the two superpowers all but impossible.
Meanwhile, the focus on national-security risks distracts from the more immediate threats of mass unemployment and rampant cyberattacks. The quest for AI supremacy also undermines the trust needed for countries to work together to tackle climate change, pandemics, conflicts, and other pressing shared challenges.
The countries with the most to lose from the AI race may be the ones not even participating in it. In its World Development Report 2026, the World Bank argues that AI offers developing countries “a rare path to greater prosperity,” pointing to their ability to leapfrog older technologies, as they did with mobile phones and distributed renewable energy. Digital connectivity has made fixed-line phones redundant, while fintech and digital-health solutions have given millions of poor people access to services at a fraction of the cost of traditional providers.
But the same logic does not apply to AI, which is eliminating the low-cost, repetitive jobs that have historically provided the bottom rungs of the development ladder in Europe, the Americas, and Asia. Robots are replacing people on production lines, while cloud-based chatbots are threatening the call centers that employ more than 17 million people worldwide.
This is troubling enough for advanced economies and China, where workforces are shrinking due to low fertility. But Africa needs to create 21 million new jobs every year just to keep pace with the continent’s growing population. With aid budgets drying up and climate change wreaking havoc, Sub-Saharan Africa is already 3.4 percent poorer than it was a decade ago. Instead of a “rare path” for growth and development, AI will make it even harder for the world’s poorest countries to catch up.
The future won’t wait
Although AI may deepen divisions within and between countries, that is no reason to abandon it. On the contrary, it is already enabling extraordinary breakthroughs in medical research. It is, for example, cutting the timelines for the development of new drugs, including treatments for Parkinson’s and many rare diseases AI itself has helped discover.
But we need to ensure that AI serves all of humanity, not just a small minority. Past technological revolutions offer some lessons here. The Industrial Revolution eventually raised the incomes of British workers, but it took a century for most people to see improvements in their living standards. Meanwhile, life expectancy fell amid pollution and dangerous working conditions in the “dark satanic mills,” and the British Army was mobilized time and again to quell labor protests. The same technological advances also facilitated the industrialization of slavery and fueled imperial conquest.
Four centuries earlier, the invention of the printing press ushered in the Renaissance, bringing extraordinary technological and artistic breakthroughs. But the technology that spread scientific ideas and humanism also allowed the Italian friar Girolamo Savonarola, inventor of the political pamphlet, to attack intellectual freedom and rally support for his bonfire of the vanities. Martin Luther used the printing press to challenge the hegemony of the Catholic Church, inadvertently setting off centuries of religious conflict that suppressed the diversity of ideas and open debate that had been at the heart of the Renaissance. It took another 160 years for the Enlightenment to take hold.
AI is just as transformative, but it is moving much faster. The Industrial Revolution gave governments a century to improve working conditions and ensure that prosperity reached most people. This time, we won’t have nearly as long to put things right.
Ian Goldin, professor of globalization and development at the University of Oxford, is the author of “The Shortest History of Migration” (Old Street Publishing, 2024, “The Experiment,” 2025) and co-author (with Tom Lee-Devlin) of “Age of the City: Why Our Future will be Won or Lost Together” (Bloomsbury Continuum, 2023). This article was distributed by Project Syndicate.