AI Could Turn Democracies Into Petrostates

This is current petrostate reality. It could also be our AI future.

Whether that future materializes hinges on how the gains from AI are distributed. If AI’s productivity-enhancing benefits are widely shared, it could deliver higher wages, shorter working hours, and broad-based prosperity that reinforces democratic institutions. If, however, the benefits accrue narrowly to its owners in an oligopolistic market structure, political and economic outcomes may begin to resemble those of hydrocarbon-dependent states such as Qatar, Saudi Arabia, or Russia.

At first glance, the political economy of oil and of AI would seem completely different. AI is bleeding-edge technology; oil has been with us in an industrial sense since the 19th century. AI conjures images of modern-day centaurs—human ingenuity wedded to AI’s problem-solving horsepower. Oil conjures images of roughnecks toiling in the Texas sun. Oil is a largely substitutable industrial input; AI is a general-purpose technology. Oil markets are dominated by huge multinationals and national firms; AI is creating a huge start-up boom.

These dissimilarities make their commonalities easy to miss. Both have high start-up and R&D costs, implying an eventual marketplace dominated by a small number of large firms. These firms are (in the case of oil) and will be (in the case of AI) capable of generating massive economic rents relative to the size of their labor force. In the case of AI combined with roboticization, capital will become increasingly substitutable for labor across an increasing number of sectors and roles.

The political consequences may be similar: an economy in which the state derives significant revenue from a narrow, high-rent sector, while the relative economic leverage of labor declines. Historically, such configurations have been associated with weaker pressures for democratic accountability—which bodes poorly for a host of important economic and social outcomes.

At the same time, the late 18th and early 19th centuries were marked by rapid productivity growth and intense geopolitical competition. Trade remained limited, and mercantilism and territorial conquest still defined the global economy and policy orientations of the nascent major powers. In this environment, states required large standing armies and ever-greater fiscal capacity. Capital stocks expanded rapidly, but so too did the demand for labor in factories and armies, elevating the bargaining position of the working class.

The confluence of these two developments—the increasing complexity and breadth of the tax base and the rise in the relative bargaining power of labor—helped give rise to more democratic institutions. Subsequently, those institutions created incentives for governments to invest in redistributive policies that helped maintain social cohesion but also in expanding the quality of their labor pools, both in terms of health and education. This was not noblesse oblige: It resulted from a combination of national competitiveness policy and the cost of ruling a populace with genuine leverage over policy outcomes. Where sovereigns required revenue from wide, mobile, and economically productive populations, taxation begat bargaining, and bargaining begat democratic representation.

These processes have not occurred in petrostates. In petrostates, enormous, easily taxable rents are generated by sectors with minimal employment relative to output. Take Qatar, where oil and gas accounted for about 40 percent of GDP and 80 percent of government revenue but at most 2 percent of the workforce—and much of that labor is foreign. Because Qatar’s energy rents generate a massive surplus relative to its tiny population (about 3 million people), the surplus supports large social welfare expenditures and high standards of living—albeit with no citizen voice in government. Most manual and service labor is performed by migrants with highly circumscribed political and economic rights, leaving the citizen population largely detached from many productive activities. Qatar represents one petrostate equilibrium: one of generalized abundance without politically consequential citizen labor or say in governance.

The politics are inherently more complicated and messier as energy rents remain the base of government revenue, but the rents must be spread over a much larger populace. In these circumstances, energy rents continue to insulate governments from pressure to democratize but fail to provide generalized abundance—the vast majority of citizens must work to avoid living at the margins of subsistence while an elite, rentier class is able to live in opulence. Young people are overeducated relative to the job opportunities that await them, creating frustrations about upward mobility and the ability to provide for a family. Political fights over the distribution of energy rents can be fierce, and governance challenges are increasingly likely to be met with force rather than accommodation. Though different in degree, these conditions roughly describe present-day Saudi Arabia and Russia.

The parallels to AI are structural. Like hydrocarbons, advanced AI is extremely capital-intensive to develop but offers high returns to scale. But to a much greater degree than energy exports—which primarily affect domestic labor markets by causing the currency to appreciate and rendering manufactures less competitive—AI may destroy many existing jobs because its scaling acts as a substitute for cognitive and (paired with roboticization) physical labor. This is where the analogy breaks down, but the implications augur even more in favor of weakened bargaining power of labor. Oil displaces labor indirectly, via exchange rates and resource pulls. AI is already automating many of the tasks previously occupied by early-career white-collar workers, creating concerns about the labor market’s ability to absorb highly educated workers, and AI plus expanded roboticization will increase pressure on manufacturing and trade-based labor as well.

In oil economies, oil prices determine rents, so boom-and-bust cycles are common. With AI, the dynamics should be different. But if AI begins displacing labor to a significant degree, it could decouple labor-market dynamics from the business cycle, leading to a labor market that looks frozen in “bust” conditions even in GDP “boom” times. Oil may provide fungible resources for governments to invest in surveillance and repression, but AI itself directly enables surveillance activities, a quality on which both authoritarian and democratic governments are already seeking to capitalize.

Automated facial recognition, predictive policing, and algorithmic content moderation make opposition movements easier to detect and suppress at scale. In authoritarian East Germany, running an analog surveillance state was costly and employed as much as an estimated 1 percent to 2 percent of the population on at least a part-time basis. Surveillance was labor-intensive and costly. AI could significantly reduce the costs and enhance the effectiveness of surveillance and repression, weakening the ability of society to organize and further eroding citizens’ bargaining power. Indeed, the combination of these two channels—eroding power of labor and decreasing costs of repression—underpins Princeton political economist Carles Boix’s arguments about why AI threatens democracy.

There is, of course, the potential for AI to become commoditized, following the path of other inventions like the internet or the computer that are considered general-purpose technologies. Currently, the main revenue driver for the AI industry is the chip sector; rents from the actual consumer-facing product—the models themselves—are being driven down via competition between the few companies operating frontier models. If these conditions continue, rents outside the chip sector would dissipate and these governance concerns would not materialize. That may happen. Even if it does, commodity markets are not rent-free—oil is itself a commodity. The question is not whether rents exist, but who captures them and how concentrated they become. And this is where the economies of scale have teeth.

An aerial view of a data center in Vernon, California, on Oct. 20, 2025. Mario Tama/Getty Images

AI scale economies are extreme because of the massive compute necessary to build frontier models and vertical integration, with AI giants not just controlling the technology but also the data centers and some of the chipsets that power the models. The conditions that result in what economists call “Bertrand competition” and rents dissipating—perfectly substitutable products with low barriers to entry—are not present. The barriers to entry at the technological frontier are punishing. The relevant comparison remains oil, not iron or wheat.

This could be the AI future: A small number of firms will generate an increasingly large share of GDP and government revenue from smaller and smaller workforces. The tax base will narrow not because of a collapse in output but because that output is increasingly dominated by the AI firms themselves, rather than laborers whose labor is taxed and whose consumption is taxed. In a recent National Bureau of Economic Research paper, Anton Korinek and Lee Lockwood argue that transformative AI could erode both of the tax bases—labor income and human consumption—that underpin modern fiscal systems, raising fundamental questions about the sustainability of the redistributive mechanisms on which democratic bargains depend. The desire for AI firms to increasingly compete directly with other knowledge-producing sectors of the economy—pharmaceuticals, law—is a harbinger of this. As automated drones and robots displace humans from the battlefield, they will further detach the waging of war from generalized costs on the populace.

With education and worker health less and less significant contributors to both output and revenue, expenditures on health care and education will decrease or shift toward maintaining base social stability rather than promoting intellectual and physical development. In short, AI economies could begin looking more and more like petrostates.

This outcome is not foreordained, but there are strong historical political economy reasons to treat it as a likely case, rather than a dystopian tail risk. Democracy is hard to sustain. It is harder still when most citizens lack the economic leverage—through labor, taxation, or military service—that historically secured them a voice in governance.

Of course, the rents generated by AI largely accrue to private firms, rather than the government. This would seemingly augur against resource curse-like dynamics, as the state still must extract revenue rather than having it flow directly into government coffers. While this is true, it is bounded. Governments tend to provide a mixture of public and private benefits that reflects the political preferences of their tax bases, assuming those tax bases are not effectively captured.

With oil, the fixity of the productive assets means capture is always possible. Even if not captured, the implication is that an AI-centric tax base would yield AI-friendly policies, with potential for regulatory capture that could lock in incumbents. And there are plenty of cases—Nigeria, Equatorial Guinea, and Russia prior to its sector’s renationalization—where private ownership or public-private partnerships coexisted with cursed dynamics. Private ownership may be on margin beneficial in weakly institutionalized contexts, but it is no panacea.

A girl jumps into the Oslofjord in Norway’s capital on July 16, 2020.Odd Andersen/AFP via Getty Images

Second, can mechanisms be put in place to ensure that whatever rents are generated are dissipated through society via social transfers, helping solve the problem of rent concentration? The stability of democratic institutions depends less on the level of prosperity than on how broadly the sources of that prosperity are distributed.

The answer to this question flows more straightforwardly from the literature on the resource curse: If an economy already had broadly representative democratic institutions and social safety nets prior to resource boom, the answer is yes; think Norway’s experience with oil. It has used its sovereign wealth fund, anchoring oil-based spending to that fund’s expected real return (3 percent annually, though the real rate of return has consistently doubled that), and coordinated wage bargaining to minimize oil rents’ distortions to Norway’s broader economy. Korinek and Lockwood identify several of these mechanisms as potential solutions to the narrowing tax base problem, along with AI-specific mechanisms like taxes on compute and roboticization. But as with oil-rich states, whether these policy levers are pulled is one of political will and incentives. And not all countries are equally positioned to make these choices.

Norway was a low-inequality, consolidated democracy when oil rents came online, with relatively limited inequalities across ethnic, regional, or sectarian lines. Where social divisions are less stark, redistributive bargains are easier to sustain and public-goods provision is more politically viable. Norway represents the best-case institutional sequencing for turning the oil curse into a blessing: strong, inclusive institutions before resource rents, rather than the other way around. If AI rents arrive in already unequal and polarized societies, political incentives will likely favor targeted protection and elite capture rather than universal redistribution.

The challenge is not that AI will make societies poorer. Rather, it is that it will make them richer in ways that render many citizens economically dispensable for continued economic growth. Bargaining between states and societies over revenue and public services has been the bedrock of durable democracies. If AI concentrates rents and reduces the bargaining leverage of labor, that bedrock will erode, and our AI future will increasingly resemble the petrostate present.