How to Survive the AI Shock
The dawning age of artificial intelligence holds great promise for the world economy and for the United States. Like so many other countries, the United States has endured decades of slow growth in labor productivity. Productivity, the amount of goods and services produced per worker, is the single most important measure of a country’s overall economic success. Slow productivity growth has meant poor growth in average incomes, which in turn has fueled much of the political turmoil manifested across the globe in the last generation. In the United States, slow growth in productivity has contributed to escalating political polarization and what many see as the gradual death of the American dream.
The United States needs to conjure up a productivity renaissance. And AI seems to be a perfect catalyst. In 2024, U.S. productivity in the nonfarm business sector grew by three percent, the fastest increase in a nonrecession, nonpandemic year in decades. A McKinsey study published in 2025 estimated that by 2030, AI-powered agents and robots could generate somewhere between $2.9 trillion and $6.4 trillion in new annual economic value for the United States—a productivity gain equal to nine to 20 percent of the country’s 2025 gross domestic product—and $28.7 trillion for the world overall.
AI is poised to power this kind of renaissance because it drives both capital investment and innovation, the two ways that countries can reliably increase their productivity. The United States and a growing number of other countries are witnessing a surge in capital investment in AI infrastructure, including computer chips, data centers, and electrical systems. According to the Federal Reserve Bank of St. Louis, AI-related investment accounted for about 39 percent of total U.S. GDP growth in 2025. That number is set to go up in 2026. According to the Financial Times, the four largest U.S. “hyperscalers” in AI and cloud computing—Alphabet, Amazon, Meta, and Microsoft—plan to increase their collective capital spending on AI in 2026 to $725 billion, a 77 percent increase from the roughly $410 billion they collectively spent last year.
The other means of increasing productivity, innovation, involves discovering new goods and services and more efficient ways to produce existing ones. Every day, companies are using AI to discover and take advantage of efficiencies and automate routine tasks. For example, AlphaFold, the AI program developed by DeepMind, whose scientists won the 2024 Nobel Prize in Chemistry, has dramatically quickened the discovery and analysis of the proteins needed to produce breakthrough drugs and other medical advances.
Yet despite all this economic promise, AI is also creating great political peril: the potential destruction of jobs at a faster rate and a wider scope than existing government assistance can reasonably address. This conundrum is hardly new. Innovation has almost always reduced labor demand for some goods and services. Sometimes that innovation did not result in widespread loss of existing jobs, but rather in the disappearance of old tasks and the invention of new tasks and capabilities within those occupations. In the early 1980s, for instance, around 16.5 million Americans worked in office operations and administrative support. Despite all the information technology inventions that have automated such secretarial tasks as taking dictation and making copies, almost the same number of Americans work in this sector today. They just do different things.
Sometimes, innovation permanently reduces demand for certain jobs: craftspeople producing horse-drawn wagons when the automobile was invented, typists with typewriters at the dawn of personal computers, workers on assembly lines automated with robots and other capital investments. This destruction is a fundamental feature of productivity growth. Indeed, the destruction of existing jobs, companies, and sometimes entire industries is a necessary condition for creating new jobs, higher incomes, and greater wealth.
But creative destruction also harms some workers and communities and can spark political resistance. This dynamic has been true since at least the Industrial Revolution, when the Luddites, skilled British textile workers protesting the invention and implementation of new machines, raided factories and destroyed power looms and knitting frames. Centuries later, the flood of low-cost Chinese exports into the United States reduced demand for many American-made goods, undercutting thousands of U.S. manufacturers and gutting many communities.
Today, evidence is building that AI is reducing labor demand in many industries, leading to an “AI shock” akin to the “China shock” of the early twenty-first century. But whereas the China shock was mostly confined to older workers in a few industries, the coming AI shock may ultimately prove much larger and more destructive. It is predominantly affecting the young rather than the old, the more educated rather than the less educated, and the full sweep of industries rather than mainly manufacturing. And because of the breakneck pace of innovation, the AI shock is reverberating much more rapidly than the China shock did.
AI is creating great political peril.
If the scope and speed of the AI shock exceed the capacity of policymakers to find solutions that blunt its negative effects, the repercussions may be severe. Countries that fail to institute adequate labor-market supports for displaced workers may lose out on the productivity gains of artificial intelligence if they bow to public pressure to pass new laws and regulations that stifle or even reverse its spread. They may face political turmoil along new, sharper cleavages. And they may fall permanently behind countries that manage to mitigate the AI shock and thus realize its full gains.
Some companies have taken welcome steps to help affected workers: Jamie Dimon, the chief executive and chair of JPMorgan Chase, and Dan Schulman, the CEO of Verizon, have each announced programs to assist employees whose jobs have been or will be eliminated because of AI. But private-sector solutions are insufficient to address AI’s economy-wide effects. Responding to the AI shock effectively requires meaningful public investment now. The United States is singularly unprepared for such an undertaking. Its antiquated labor-market policies are narrowly focused and inadequately funded. Without new sources of tax revenue to support and retrain displaced workers, the country will simply not be able to cope with the scale of the shock.
A pair of new policies can prevent the United States from repeating the mistakes of the China shock: tax credits to encourage training in new skills, and wage-loss insurance to encourage reemployment. These new fiscal outlays should be funded by a new tax on equity-related compensation, which would more fairly distribute the windfalls of AI’s expected productivity gains among the executives of companies that will financially benefit most from the technology’s disruptive effects and the workers hit hardest by it. The AI boom will assuredly transform the U.S. economy in ways no economist can fully anticipate. But these policies will ensure that its gains are more broadly shared across innovative companies, their workers, and communities across America.
DEGREES OF DIFFICULTY
Consider the following headlines from late 2025. In October, Amazon announced the elimination of 14,000 jobs along with plans for further cuts. A month later, Verizon eliminated over 13,000 jobs—roughly 13 percent of its workforce and its largest round of layoffs ever. According to the executive coaching firm Challenger, Gray & Christmas, job cuts announced across the United States in 2025 totaled 1.2 million, a 58 percent increase from the previous year and the highest tally, excluding the COVID-19 pandemic period, since the nadir of the Great Recession in 2009.
Increasingly, companies cite AI as the reason for such layoffs. In June 2025, Amazon’s CEO, Andy Jassy, warned that the company expected to “reduce our total corporate workforce as we get efficiency gains from using AI extensively” over the next few years. In November, Enrique Lores, then the head of HP, explained a cut of approximately 5,000 jobs as part of the company’s effort to embed AI “in everything we do.”
This year has brought more of the same. In January, Bank of America’s CEO, Brian Moynihan, predicted staff reductions as the company pursues “operational excellence and application of new technologies, including AI.” The following month, the financial services company Block announced that it would lay off nearly 40 percent of its workforce—over 4,000 employees—thanks to the adoption of AI “intelligence tools.” In April, Meta announced plans to lay off about 8,000 employees—ten percent of its workforce—and scrap 6,000 open positions while expanding its AI investments. The same day, Moynihan revealed that AI had already eliminated 1,000 jobs at Bank of America. In May, the chief executive and president of Cloudflare justified the elimination of 1,100 jobs—not as a “cost-cutting exercise” or “an assessment of individuals’ performance,” but as an effort to define “how a world-class, high-growth company operates and creates value in the agentic AI era.” Not all these recent layoffs can be attributed solely to AI, but employee efficiency is at the heart of many decisions to deploy the technology.
The question is not whether AI will bring forth a productivity bonanza; it will.
Although the growing volume of AI-related layoffs means that this emerging shock is affecting many kinds of workers, it seems to be disproportionately hitting the young and the more educated, across all industries. One of the most definitive academic studies to date, by the Stanford Digital Economy Lab, examined monthly payroll records for millions of people across tens of thousands of companies from the months before the initial public release of OpenAI’s ChatGPT, in November 2022, through July 2025. Drawing on data from ADP, the largest payroll software provider in the United States, the researchers found that workers between the ages of 22 and 25 in the occupations most exposed to AI, such as software developers and customer service representatives, experienced a six percent decline in employment. “In contrast, employment trends for more-experienced workers in the same occupations, and workers of all ages in less-exposed occupations such as nursing aides, have remained stable or continued to grow,” the paper said.
The China shock brought blue-collar, less educated American workers into direct competition with hundreds of millions of low-wage Chinese workers. In contrast, a Federal Reserve Bank of New York study concluded that recent increases in overall U.S. unemployment were concentrated “among recent college graduates and white-collar workers.” Across the entire U.S. economy, the unemployment rate for recent college graduates is outpacing that of the overall labor force. In March 2026, the jobless rate for recent college graduates stood at about 5.6 percent, compared with 4.2 percent for the overall population. Perhaps the most prominent early casualty of AI’s pressure on the white-collar job market is software engineers, particularly recent STEM college graduates struggling to secure entry-level work as AI’s ability to write and edit computer code improves by the day.
But the effects are hardly confined to aspiring coders. The St. Louis Fed study concluded that “even highly educated workers in previously stable fields are not immune to economic disruption.” In many previously shock-resistant industries today, people whose main tasks can be largely if not entirely codified by a set of rules and processes—and thus can be automated away by AI—are now under threat: analysts in consulting, accountants and actuaries in finance, associates and paralegals in law. This slackening of labor demand for the young reduces the likelihood that early-career workers will acquire the differentiated skills that tend to catapult their future career prospects.
In some industries, particularly those that rely on so-called tacit rather than codifiable knowledge and for which aptitudes such as judgment and emotional intelligence matter, AI is augmenting work, not automating it. One example is health care. The New York Times reported that the Mayo Clinic has expanded its radiology staff by more than 50 percent even while deploying hundreds of AI models to support image analysis, allowing doctors to focus more on complex decision-making and patient care. AI is also spurring the creation of altogether new occupations and someday may even spawn new industries.
AI should eventually boost productivity, incomes, and the overall standard of living in the United States. But even as AI is beginning to complement existing labor, the relative speed and scale of initial job destruction are likely to exceed the speed and scale of job creation. And the damage it wreaks may be exponential, as well. One reason is the astonishing rate of improvement in the quality of AI tools. Since the release of ChatGPT in late 2022, the number of AI offerings has exploded and their caliber has exponentially improved—all at little cost to any business or individual with an Internet connection. This year, in March and April alone, at least ten new models were released by the leading AI companies.
History offers few, if any, examples of such prodigious innovation, adopted so rapidly. A U.S. Census Bureau survey published in 2024 reported that 60 years after the first industrial robot was installed in the United States in 1961, only 12 percent of manufacturing factories used robots. Another found that by 1997, a generation after the advent of personal computers, only 49.8 percent of employed adults used them. According to the St. Louis Fed’s November 2025 study of AI adoption, 54.6 percent of U.S. adults were using AI—just three years after the public launch of ChatGPT.
Business leaders are already bracing for the broadening impact of the AI shock. Forty-one percent of approximately 10,000 global executives polled in a 2024 World Economic Forum survey said they expected AI to reduce their workforces by 2030. They are no doubt aware of the emerging scholarly consensus on AI’s labor-market effects. Using a novel methodology that modeled 32,000 distinct skills across the U.S. labor market, a team of scholars at the Massachusetts Institute of Technology calculated that 12 percent of total U.S. wages pay for tasks that are technically automatable by current AI systems—calling the early effects on tech-related jobs like coding “only the tip of the iceberg.” The McKinsey study concluded that current technologies, including AI agents, “could, in theory, automate activities accounting for about 57 percent of U.S. work hours today”—in other words, “a majority of the work now done by people in the United States.”
THE YOUNG AND THE RESTLESS
The question is not whether AI will bring forth a productivity bonanza; it will. The question is whether the United States can fully realize those future gains by navigating the initial disruptions to labor markets and mitigating the strains on the country’s economic and political structures. The historical record is unambiguous: resistance to innovation arises when the pace and scope of creative destruction exceed the pace and scope of a government’s policy response.
So far, the largest creative destruction shock to the post–World War II global economic system was the explosion of labor-intensive Chinese exports resulting from the productivity boom that began in China in the early 1980s. Studies have shown that from the 1990s through the first decade of the twenty-first century, U.S. imports of Chinese goods destroyed more than a million U.S. manufacturing jobs, with many of the losses concentrated among older manufacturing workers in a small number of states with once thriving industrial economies, such as Ohio and Pennsylvania.
That Washington and local governments did too little to address the economic, cultural, and social costs of the China shock has been well documented. They incorrectly assumed that the larger but diffuse benefits—cheaper consumer goods, expanded export opportunities—would outweigh the concentrated and deep harm of job loss in the eyes of displaced workers. The main U.S. policy bulwark against the China shock was the Trade Adjustment Assistance program, which was established under the Trade Expansion Act of 1962 as part of that bill’s efforts to liberalize U.S. trade under the General Agreement on Tariffs and Trade. The TAA aimed to provide expanded unemployment insurance, job-search assistance, and retraining to workers whose companies had been harmed by increased imports. But the TAA was hobbled by its complex design, cumbersome bureaucracy, and inadequate funding: in 2005, its total outlay was $845 million, just 0.03 percent of total federal spending.
The ground is fertile for an AI backlash.
The social consequences of this neglect were devastating. The workers most hurt by the contraction of the manufacturing labor market were far more likely to suffer what the economists Anne Case and Angus Deaton have described as “deaths of despair”: suicide, drug overdoses, and alcohol poisoning, an explosion of which has contributed to declines in life expectancy among some groups of Americans over the last two decades. The political consequences were similarly seismic. Donald Trump was elected and reelected president largely because of his strident rejection of globalization and his willingness to give voice to the grievances of China shock victims. In his first term, he launched a trade war against China and pursued immigration restrictions. In his second, he instituted a sweeping and chaotic global tariff regime and a mass deportation campaign focusing primarily on immigrants working low-wage jobs in the United States illegally.
These shock waves continue to define American politics today despite two important and underacknowledged realities. The economic pressures from China on less-skilled manufacturing jobs have largely eased, and those pressures were outweighed for the country overall by lower prices of consumer goods and industrial inputs, as well as growth in sectors in which the United States retains a comparative advantage over the rest of the world. In 2024, the median U.S. household income was $83,730—18.6 percent higher in inflation-adjusted terms than it was in 2001, the year China acceded to the World Trade Organization—and aggregate unemployment remains near historic lows. But these economic realities have not softened the political backlash against globalization.
The China shock entailed a reorganization of global supply networks, its pace dictated by the years-long crawl of World Trade Organization agreements and multinational corporate decision-making. According to a canonical study by the economists David Autor, David Dorn, and Gordon Hanson, it caused the loss of around 1.5 million U.S. manufacturing jobs between 1990 and 2007, displacing an average of about 7,500 U.S. manufacturing workers every month. The evidence suggests that the AI shock may already be as large, with the potential to grow much bigger. In a U.S. economy that contains two million software-developer jobs and three million customer-service jobs alone, the number of jobs threatened by AI could dwarf the number of jobs lost to the China shock. Moreover, the digital infrastructure of AI is vastly easier to build than the physical infrastructure of global supply networks—or of earlier epoch-defining innovations such as railroads or the electric grid. The AI shock may end up being the fastest technology-driven disruption in human history.
The ground is already fertile for a political movement animated by backlash to the AI revolution. Younger, more highly educated workers around the world are already voicing profound dissatisfaction with the economic, political, and social systems into which they are maturing. The outlook of Generation Z—skeptical of elites, capitalism, and Big Tech—has been deeply shaped by pessimism over stagnant wages, concentrated wealth, fading opportunity, and declining trust in institutions and their leaders.
This generation is entering the political fray at a time when the median age of a first-time homebuyer is 40, up from 32 in 2000 and 28 in 1991; the monthly premiums for high-quality health care are surging; and the annual sticker price for undergraduates at top universities is approaching $100,000. It is not surprising, then, that today most young Americans believe that capitalism does not provide fair opportunities to succeed, or that only 16 percent of Americans under 30 believe that democracy is working well for them. Accordingly, they are supporting candidates who speak to these concerns: Abigail Spanberger and Mikie Sherrill, both of whom campaigned on “affordability agendas,” rode the youth vote to respective victories in the Virginia and New Jersey gubernatorial races, and Zohran Mamdani, a democratic socialist, galvanized young voters with a similar message en route to becoming mayor of New York City.
In many advanced countries, much of Generation Z sees its progression into adulthood as a gauntlet of seemingly endless competitions: for admission into selective middle schools, high schools, and universities; for hiring by selective global companies for selective internships; and, finally, for permanent employment by those companies to launch their careers. If AI eliminates too many of these coveted jobs, the sense of betrayal among young, highly educated professionals may closely echo the outrage of older, less educated manufacturing workers during the China shock. As age becomes another fault line in American politics, new battles over Social Security and Medicare could further divide an already polarized country.
UNLEARN HOW TO CODE
Policymakers who assume Americans will uncritically embrace AI, labor-market disruptions and all, will be in for a rude awakening. Indeed, before it has begun in earnest, the AI transition is already unpopular. A September 2025 Marist poll revealed that roughly 67 percent of all Americans believe that AI will eliminate more jobs than it creates, and a March 2026 Quinnipiac University poll found that 81 percent of young Americans believe that it will reduce job opportunities. The backlash against data centers is the first surge in a building groundswell of discontent. Investment in the data centers that provide the computing infrastructure needed for the broad deployment of AI is increasingly driving U.S. economic growth. That has not stopped the proliferation of heated protests against data centers or of proposed moratoriums on their construction being debated in statehouses and town halls across the country.
To reap the full productivity and geopolitical gains from AI and avoid a wider backlash that could seriously limit those gains, the U.S. government will need to devise new ways of supporting workers. Tax credits to encourage training in new skills and wage-loss insurance to stimulate reemployment are sufficiently broad, simple policies that will bolster skills and incomes, help workers navigate AI-related job losses, and thus avoid another moment of political rupture. To pay for these policies, the federal government should establish a payroll tax on companies’ equity compensation.
Most employees are paid for the work they do entirely in cash that can immediately be taxed by the federal government. But some, particularly the country’s highest-paid executives, receive most of their compensation in the form of equity incentives, such as shares in company stock or options to purchase those shares. Equity compensation is generally not taxed at the time it is granted or as it compounds in value over time. According to the Economic Policy Institute, equity compensation has been the single biggest driver of inequality of pay between American CEOs and their average workers, which has risen from a ratio of 21 to 1 in 1965 to 281 to 1 in 2024.
Whether explicitly or implicitly, markets currently reward companies for paying equity compensation. Under U.S. law, public companies must report their financial results in accordance with Generally Accepted Accounting Principles. But most companies also report “pro forma” results that exclude stock-based compensation. Academics and market observers routinely criticize this practice because pro forma results overstate corporate profitability by excluding the cost of equity compensation. Investors tolerate the convention, however, so companies continue to do it. This practice greatly rewards executives who get paid mainly in equity and disadvantages workers who receive cash.
The AI shock may already be as large as the China shock.
A new payroll tax of 25 percent levied on public and large private companies when issuing equity compensation would generate at least $100 billion in annual revenue without unduly penalizing AI innovation. It would link the generated tax revenues to the appreciating equity of the companies that benefit most from an AI-related productivity boom, while still allowing executives to earn compensation packages consistent with market forces. Those companies’ higher revenues and lower costs, made possible by the successful deployment of AI, would offset some, if not all, of the tax’s cost over time. The $100 billion in new annual tax revenue would be equal to just 0.16 percent of the equity-market capitalization of the Fortune 500 companies as of May 2026 and only about ten percent of the $1 trillion in U.S. stock repurchases in 2025.
Unlike a wealth tax, which may be unconstitutional and could squelch extraordinary innovations like AI, a payroll tax on equity compensation would align the economic interests of all those with a stake in artificial intelligence: companies looking to innovate and grow, a federal government looking to boost overall U.S. productivity and average living standards, and workers who need support to navigate the AI transition.
The new revenue would fund both the tax credits to encourage retraining and the wage-loss insurance to encourage reemployment. To cushion the blow of potential mass white-collar job losses, the federal government would create a substantial new tax credit that eligible workers or their employers could use to invest in building new skills. The government would not be creating its own retraining programs and thus would not be forced to predict which skills will be most in demand; the training, including online courses, in-person classes at colleges and universities, and in-house programs crafted by companies, would teach whichever skills end up being a strong complement to AI.
A widely available portable tax credit would also overcome the so-called adverse-selection problem that befalls most training programs, which generally do not boost worker earnings because companies assume that those who participate in them may somehow be less skilled and thus less desirable to hire. By offering this tax credit to many workers rather than few, competition would compel companies to offer meaningful retraining or risk seeming unattractive in the overall labor market.
To encourage the reemployment of workers who lose their jobs to AI, the new wage-loss insurance program, also created by the federal government, would, for a time, replace a fraction of eligible workers’ lost wages once they have found another, lower-paying job.
The rationale for wage-loss insurance is straightforward. Laid-off workers often lose firm- or industry-specific capital; for example, software engineers at a large tech company laid off because AI has automated away their roles miss the chance to gain experience using AI in the workplace. Often, these workers can find jobs only at a lower wage, which they are understandably reluctant to take, thus aggravating the loss of human capital by remaining unemployed. The government can mitigate this risk by partially compensating workers for their lost earnings when they move to lower-paying jobs. A software engineer, for example, might find a new job in sales at another large tech company or as the chief engineer of a fledgling AI startup. Wage-loss insurance would help soften the blow of accepting a lower salary in the new position.
Wage-loss insurance incentivizes faster reemployment after a job loss by reducing the worker’s incentive to stay unemployed in the hope of landing a better job. And it does so without discouraging businesses from hiring because the insurance does not alter the market wage firms are willing to pay.
The emerging evidence that AI can substitute for high-talent tasks (and thus for high-wage jobs) makes wage-loss insurance especially well suited to addressing the perils of the AI shock—and its promise, as well. Research by the economists Rob Shimer and Daron Acemoglu has shown that wage-loss insurance increases aggregate output by encouraging workers to seek high-wage jobs that carry high unemployment risk along with high productivity potential—in other words, the very jobs that will power an economic boom.
The relative simplicity of tax credits and wage-loss insurance would make adjusting their eligibility criteria and overall scope simple, as well. For example, the government could easily change or expand the list of eligible occupations and skill groups to keep pace with the AI shock as it unfolds.
Creative destruction has been central to growth in the world economy for centuries. Artificial intelligence is no exception to this rule. As a technology, it holds sorely needed economic promise for workers, companies, and the country. But if not managed properly, its adoption could yield a political and social crisis that could make the China shock look mild by comparison. Acting boldly now, before the AI shock has fully materialized, can forestall another era-defining crisis and usher in a productivity renaissance.