AI or Nuclear War: The World at a Crossroads

The article compares economic inequality in 1928 and 2024 (an identical Gini coefficient of 0.49) and analyzes two opposing scenarios for the future: the transformative potential of artificial intelligence as a tool for addressing civilizational challenges versus the escalation of geopolitical conflicts toward a nuclear confrontation. Drawing on historical parallels, it shows that technological revolutions on their own do not resolve inequality without political will.

In 1928, the Gini coefficient of American households reached 0.49. The top one percent of the population took home nearly a quarter of all income. A speculative bubble on Wall Street was inflating stock prices to absurd heights, while workers' real wages stagnated. A year later came the crash, the Great Depression, and a decade that reshaped the face of American society.

In 2024, the Gini coefficient of American households reached 0.49. The top one percent of the population takes home 21% of all income. Asset prices — stocks, real estate, cryptocurrencies — are at historic highs, while the bottom half of the population has seen no real gains since 1978. The ratio of CEO pay at S&P 500 companies to the median wage of their employees has reached 285:1 (AFL-CIO Executive Paywatch, 2025).

The numbers are virtually identical. But the situation is fundamentally different. In 1928 there existed mechanisms that societies had repeatedly used to correct extreme inequality: great wars, revolutions, pandemics, the collapse of states. The historian Walter Scheidel calls them "the four horsemen of leveling." Today — as we will show in this article on the basis of a synthesis of dozens of studies, from cliodynamics to nuclear strategy — all four have been systematically taken out of service. War in the nuclear age threatens existential catastrophe. Pandemics in the era of modern medicine and quantitative easing paradoxically deepen inequality. Revolution and collapse run up against the unprecedented power of the modern state armed with digital surveillance.

That leaves a single candidate for the "fifth horseman": artificial intelligence. But AI is a horseman with two heads. One promises the democratization of knowledge, a reduction in wage inequality, and the restoration of the middle class. The other erodes the nuclear balance of mutually assured destruction, compresses decision times from hours to minutes, and asymmetrically favors technologically dominant powers — thereby increasing the incentive for lagging nuclear states to reach for their weapons before they lose them.

The world does not stand at a crossroads between AI and nuclear war. It stands at a crossroads where AI can lead to both — and the outcome depends on which of its two vectors prevails first.

Peter Turchin was born in 1957 in Soviet Obninsk as the son of the dissident Valentin Turchin. In 1977 the family was expelled from the Soviet Union. Peter completed a bachelor's degree in biology at New York University and a doctorate in zoology at Duke University — where he studied the population dynamics of insects. And then he did something unexpected. The mathematical models with which he described the population explosions and collapses of beetles and butterflies, he applied to human societies. He called the result cliodynamics, and his structural-demographic theory identifies three mechanisms that together generate cyclical instability.

The first is elite overproduction — a situation in which a society produces more candidates for elite positions than it can absorb. In the modern context: a surplus of law graduates, MBA holders, and PhDs competing for a limited number of truly elite positions. Frustrated aspirants become "counter-elites" — the Gracchi brothers in Rome, Robespierre, Lenin. In the United States the number of people with assets over 10 million dollars rose tenfold in 36 years — from 66,000 in 1983 to 693,000 in 2019.

The second is the pauperization of the masses — not necessarily absolute impoverishment, but the stagnation of real wages amid rising productivity. Turchin identifies a turning point in the United States around 1978, from which wages and productivity decoupled. He calls this mechanism the "wealth pump" — the systematic transfer of resources from workers to elites.

The third is the fiscal crisis of the state — the state cannot collect sufficient taxes because the elites resist taxation, yet expenditures rise.

He tested the model on eight historical societies, from the Roman Republic to Tsarist Russia, and found quantitative confirmation in all of them. The key instrument is the Political Stress Index (PSI), composed of roughly 40 indicators — and it was on this basis that he published, in 2010, in the prestigious journal Nature, a prediction: political instability in the United States would peak around 2020.

On January 6, 2021, attackers stormed the Capitol.

But Turchin's theory is not the only one that predicted the current instability. There are at least four cyclical frameworks, each operating on a different time scale — and all of them are right now in a descending phase.

Kondratiev waves (40–60 years) describe economic cycles driven by technological revolutions. The Soviet economist Nikolai Kondratiev identified them in a series of works in the 1920s — for which he contributed to his own arrest and execution in 1938, since the theory of capitalism's cyclical renewal contradicted the Stalinist dogma of its inevitable collapse. Carlota Perez of University College London elaborated Kondratiev waves into a four-phase model: the irruption of a new technology → speculative frenzy → crisis (the bubble bursts) → a golden age of broad prosperity. The fifth wave (the IT revolution) stalled after the 2008 crisis in the transitional phase, and the "golden age" never arrived.

Wallerstein's hegemonic cycles (100–140 years) describe the rise and fall of global hegemons. The United States as the dominant power is in a phase of relative decline; China is rising, but it is unclear whether it can or wants to take over the hegemonic role. Geopolitical fragmentation — Ukraine, the Middle East, pressure around Taiwan — blocks global cooperation.

Scheidel's compression cycles have no fixed period, but they identify four historical mechanisms for reducing inequality: mass-mobilization wars, transformative revolutions, the collapse of states, and lethal pandemics. The postwar compression, which reduced inequality to historic lows, has been systematically falling apart since the 1980s.

And above it all tick Turchin's secular cycles (200–300 years), which describe the phases of expansion, stagnation, crisis, and depression of entire societies.

Four clocks on four different timepieces. All showing the same time.

The abstract mechanisms of Turchin's theory acquire sharp contours the moment we confront them with hard data on the distribution of income and wealth.

According to the World Inequality Database (WID.world), the wealthiest one percent of the world's population owns 37–45% of all global wealth (depending on the measurement methodology), while the bottom 50% of adults hold a mere 2%. Total private wealth has reached an estimated 480 trillion dollars, and a mere 56,000 people — who would fit into a football stadium — own more than the entire bottom half of humanity.

To understand the dynamics, the American figures are key. The share of the top one percent in U.S. income rose from 11% in 1980 to 21% in 2023. The share of the bottom half of the population fell from 20% to 13%. Billionaires grew at a rate of 8% per year from the 1990s, while average global wealth grew only 3.4%. And in 2024, for the first time, more billionaires acquired their wealth through inheritance than through business — a typical indicator of Turchin's late phase of accumulation, in which elites reproduce themselves as rentiers rather than productively.

Branko Milanovic of the City University of New York captured this dynamic in the famous "elephant graph," which shows who actually profited from globalization in 1988–2008. Massive gains went to the global middle class (especially China) and the global top 1%. Who lost? The lower middle class of the developed countries — precisely the demographic group that forms the voter base of Trump, Brexit, Le Pen, AfD, and all the other populist movements.

Gabriel Zucman of UC Berkeley estimated in 2015 that roughly 7.6 trillion dollars is hidden in tax havens — about 8% of global household financial wealth. Newer estimates are higher. These people are in no statistics. Real inequality is therefore systematically underestimated.

The key to understanding the current situation is Perez's model of Kondratiev waves. Every technological revolution — from the steam engine to information technology — passes through four phases. After the phase of irruption and speculative frenzy comes a crisis (typically a financial crash), after which a "golden age" follows — a period in which the benefits of the new technology spread across the whole of society. New infrastructure, new jobs, new regulation.

The fifth wave — the IT revolution — stalled. The crisis came in 2008, but the "golden age" did not arrive. Instead, central banks pumped trillions into the system through quantitative easing, which saved the financial system, but the distributional effect was directed upward, to the owners of assets. The result is an economy of "zombie firms" — according to the BIS, 12–16% of listed firms in the United States and Europe survive only thanks to cheap debt and produce no value — and a frozen transition to broad prosperity.

Why? Because the "golden age" has historically required political pressure from below — a labor movement, regulation, progressive taxation — and it is precisely this pressure that is systematically blocked by elite overproduction and the political stalemate that Turchin describes.

When all four frameworks are combined, a disquieting picture emerges of mutually reinforcing feedback loops that form a closed circle:

Financialization → inequality. Since the 1980s, the share of the financial sector in the GDP of developed economies has been rising. Financial profits make up 25–30% of all corporate profits in the United States, even though the sector employs less than 5% of the workforce. Returns on capital systematically exceed wage growth — precisely Piketty's r > g.

Inequality → elite overproduction. The rising share of wealth at the top generates more aspirants for elite positions than the system can absorb. The number of American lawyers rose by 50% over 30 years; the number of MBA graduates tripled. The frustration of unsuccessful aspirants — Turchin's "counter-elites" — feeds political extremism.

Elite overproduction → political stalemate. Elites cannot agree on reforms, because each faction protects its own rent-generating position. Political polarization in the United States has reached a historic maximum according to DW-NOMINATE scores — the overlap between the "most liberal" Republican and the "most conservative" Democrat is zero.

Political stalemate → absence of reform. Without structural reforms — progressive taxation, regulation of the financial sector, investment in wages — the Kondratiev cycle cannot move into the "golden age" phase.

Absence of reform → frozen K-cycle → social entropy → loss of state legitimacy → inability to correct.

And so the circle closes. The system needs correction, but all the mechanisms of correction are blocked — by the system itself.

Historically, four paths led out of this loop. All of them were violent. And all of them — as the following analysis will show — have today been systematically taken out of service.

The strongest argument for the leveling effect of war is the "Great Compression" — the unprecedented reduction in inequality during World War II and immediately afterward. Thomas Piketty identified four mechanisms: the physical destruction of capital (Hungary lost over 50% of its national wealth), confiscatory taxation (the United States raised the top rate to 94%), hyperinflation that decimated bondholders (the Weimar Republic: 29,500%), and land reforms (the Japanese reform redistributed 5.8 million acres). Claudia Goldin and Robert Margo called the result the "Great Compression" — the wage differential between skilled and unskilled workers fell by 35 log points. Kenneth Scheve and David Stasavage added a key mechanism: mass mobilization created a moral obligation of the "conscription of wealth" — you cannot send the poor to their deaths while simultaneously refusing to tax the rich.

But the Great Compression was a historical anomaly, not a law.

McCoy, Roscoe et al., in an article published in PNAS (2025, DOI: 10.1073/pnas.2400695121), analyzed archaeological data from 770 settlements from the Neolithic to the present and found that war both increased and decreased inequality — the deciding factor was context. In societies with less collective governance and agriculture limited by available land, war increased inequality; in early societies with more collective governance, it decreased it. The decline occurred primarily in total-mobilization conflicts — and there were minimal numbers of those throughout history. Joris Heldring reached an analogous conclusion for pre-industrial Germany (Explorations in Economic History, 2023): wars typically strengthened the position of military elites at the expense of the rest of the population.

And modern conventional wars? Joseph Stiglitz and Linda Bilmes, in their analysis of the economic costs of the Iraq War (The Three Trillion Dollar War, 2008), pointed to an unprecedented phenomenon: for the first time in American history, a war was waged on debt while taxes on the rich were simultaneously cut. Mark Peltier (Dædalus, 2025) describes how the military-industrial complex transformed war from a corrective mechanism into yet another instrument of wealth concentration: in 2024 the United States spent roughly 884 billion dollars on defense (NDAA FY2024 authorization), the bulk of which flowed to corporations whose shareholders belong to the top deciles of the income distribution.

And nuclear weapons? They do not represent a corrective mechanism — they represent an existential threat.

The relationship between inequality and war is not one-directional. John Hobson formulated, as early as 1902, the thesis that imperialism is a direct consequence of domestic inequality: the rich have nowhere to invest at home (insufficient domestic demand), and so they push for expansion overseas. Lenin took it up and radicalized it. But it was only the trio of economists — Hauner, Milanovic, and Naidu (Stone Center Working Paper, 2020) — who first tested it empirically on the data of pre-war Britain.

Their finding: inequality in pre-war Britain was at historic highs, the top one percent owned roughly 70% of wealth, and a third of the UK's net national wealth lay in foreign investment (1913). They demonstrated that risk-adjusted foreign returns did indeed exceed domestic ones, and that higher foreign assets correlated with higher war mobilization. Milanovic summed up: "All the ingredients necessary for a great conflict were present." Inequality did not produce war deterministically — but it created the structural conditions that made it more probable.

Criticism is necessary: Pseudoerasmus objects that the United States and Spain were capital importers, yet empires nonetheless. Kirshner shows that bankers actively did not want war. Ferguson documents that the majority of colonial disputes were resolved peacefully. Nevertheless, the basic logic remains substantiated: extreme inequality → a surplus of capital seeking returns → geopolitical tension.

This chain has a direct parallel in the present. Chinese investment in the Belt and Road Initiative, American technological expansionism, Russian resource imperialism — all follow Hobsonian logic: the domestic concentration of capital pushes for external expansion.

The historical mechanism of pandemic leveling was simple and brutal. The Black Death of 1347–1351 killed 30–60% of the European population. The mass loss of the workforce created an unprecedented bargaining position for surviving workers: wages rose by 50–100% within a single generation. Feudal bonds loosened. Scheidel documents this as one of the most effective corrective mechanisms in history.

COVID-19 offered itself as a natural experiment: what happens when a pandemic strikes a modern economy? The result was the opposite. Between March and December 2020, the total wealth of the world's billionaires rose by 3.9 trillion dollars. In the same period, workers globally lost 3.7 trillion dollars in income. American billionaire wealth rose by 62% — from 3 trillion to 4.8 trillion. Oxfam reports that the 10 richest people on the planet doubled their wealth. The World Inequality Report 2022 recorded the steepest increase in the billionaires' share of global wealth since 1995.

Why the opposite effect? First, mortality was too low: 0.1–0.2% of the population versus 30–60% for the Black Death. Modern medicine saved millions of lives — but at the same time it neutralized the mechanism that historically leveled inequality. Second, monetary policy: central banks responded with massive quantitative easing, which pumped liquidity into financial markets. Assets — stocks, real estate, bonds — are owned predominantly by the top deciles of the income distribution. Third, the digital economy enabled knowledge workers to work from home, while manual workers lost their jobs. Fourth, the fiscal response went through debt, not the taxation of the rich.

The pandemic horseman was neutralized by modern medicine in combination with central banking. And if a pandemic with higher mortality were to come? The same monetary and digital mechanisms would, with high probability, work again.

That leaves Scheidel's last two horsemen: revolution and state collapse. Theda Skocpol, in her seminal work States and Social Revolutions (1979), formulated a principle that the empirical evidence of the last fifteen years dramatically confirms: "The difference between a successful revolution and a rebellion is the state's ability to maintain a monopoly on the means of coercion. If the coercive apparatus remains coherent and effective, it can withstand mass discontent and survive even considerable illegitimacy."

The modern state commands instruments of control of which historical autocrats could not have dreamed. Steven Feldstein (Carnegie Endowment, 2021) documents how China combines mass camera surveillance, internet censorship, DNA collection, and AI-assisted predictive repression into an integrated system of control. Xu Xu (American Journal of Political Science) demonstrated that digital surveillance makes it possible to replace costly blanket co-optation with targeted preventive repression — to identify and isolate potential radicals before mobilization arises.

Sergei Guriev and Daniel Treisman, in their monograph Spin Dictators (Informační diktátoři) (Princeton University Press, 2022), identified a new type of authoritarianism: the "informational autocracy." Putin, Erdoğan, Orbán do not control the population through mass violence, but through the distortion of information and the simulation of democracy. The co-optation of media owners, not outright censorship. Postmodern propaganda, not totalitarian terror.

The empirical record of uprisings in 2011–2024 is devastating. The Arab Spring: of twenty countries, only Tunisia achieved a democratic transition — and in 2025 it is back in authoritarianism. Iran 2022 (the protests after the death of Mahsa Amini): suppressed by an internet shutdown and AI surveillance. Myanmar 2021: the army maintained control despite mass protests. Belarus 2020: the same. Hong Kong 2019–2020: the same.

Historical revolutions required communication failure — a period in which people could organize before the state reacted. Tiananmen in 1989 lasted weeks before the tanks came. Today the police or army intervene within minutes anywhere. The only one of Scheidel's horsemen that could theoretically function — state collapse — presupposes precisely the failure of the coercive apparatus that modern technology makes ever more improbable.

Let us summarize. Great war: nuclear weapons transformed it from a corrective mechanism into an existential threat; conventional and local wars increase inequality. Pandemic: modern medicine eliminated the demographic shock; monetary policy reversed the distributional effect. Revolution: digital surveillance enables predictive repression; "spin dictators" simulate democracy. State collapse: the unprecedented power of the modern coercive apparatus.

What remains are precisely the two scenarios we wish to avoid: either the long-term persistence of extreme inequality (feudal stability), or a global catastrophe of such scale that no state can control it.

And there is reason to believe that the first scenario has already begun.

The Greek economist Yanis Varoufakis, in his book Technofeudalism (2024), formulated a provocative but empirically grounded thesis: capitalism did not survive the 2008 crisis — it was replaced by something else. The key distinction is the predominance of rent over profit. In capitalism, profit is the reward for innovation and risk. In feudalism, profit is rent — income flowing from mere ownership. Digital platforms — Amazon, Google, Apple, Alibaba — are not markets. They are fiefdoms: they control access, set the rules, collect fees. The entrepreneurs on them are not independent capitalists — they are vassals dependent on the platforms, to which they must pay tribute.

Varoufakis calls the new ruling class "cloudalists." Data from 2024 confirm this development: for the first time in history, more billionaires acquired their wealth through inheritance than through business. Rent prevails over innovation. Capital reproduces itself instead of being created.

Stephen Kotkin, a historian at Princeton, describes an analogous process in geopolitics: the world is fragmenting into "civilizational domains" — large power blocs with their own logic, values, and ambitions. If we combine Varoufakis's economic technofeudalism with Kotkin's geopolitical fragmentation, a map of "natural neo-kingdoms" emerges:

The United States — the technological hegemony of the cloudalists, military dominance, the dollar system. An unprecedented concentration of AI capacity: in 2023 the United States alone received 67.2 billion dollars in private AI investment, 8.7 times more than China.

China — state capitalism with digital authoritarianism. A combination of market efficiency with central planning and mass surveillance. Belt and Road as neocolonial infrastructure.

India — 1.4 billion inhabitants with the caste system as a feudal substrate, a rapidly growing tech sector, a nuclear arsenal.

Russia — resource feudalism, where economic power flows from the ownership of oil and gas. A nuclear arsenal as the last guarantee of relevance.

The Persian sphere / Iran — theocratic feudalism, nuclear ambitions, control of strategic straits.

Turkey — neo-Ottoman ambitions, a geographic position between Europe and Asia.

Saudi Arabia / GCC — petro-feudal monarchies with diversification ambitions (NEOM, Vision 2030).

The European Union — a fragmented bloc, potentially the most vulnerable, because it lacks a central coercive authority comparable to the other domains.

Graham Allison of Harvard analyzed 16 historical cases in which a rising power challenged a dominant power — the so-called Thucydides Trap. Twelve of the sixteen ended in war. Of the U.S.–China relationship he wrote: "Both sides are following a script as if they were competing over who can better embody the characteristics of the rising and dominant power described by Thucydides."

Both global and local wealthy elites live in a mistaken belief, arising from an incorrect extrapolation of the past. They believe that they will increase their wealth, that they will preserve it, that they will survive. There are strong reasons to believe that none of this need be true.

First, the primary wealth of elites is internal, not external. Financial wealth — Zucman's 7.6 trillion in offshore accounts — is only one dimension. The real power of Putin, Xi Jinping, the Iranian ayatollahs, or the Indian dynasties does not rest on a bank account. It rests on control of the security forces, the courts, the media, state enterprises. This power cannot be transferred to Dubai or the Cayman Islands. Mikhail Khodorkovsky had 15 billion dollars, but not power — and ended up in Krasnoyarsk. Navalny had popularity, but not power — and died in a Russian prison above the Arctic Circle.

When an elite faces the loss of power, not merely money, it has nowhere to flee. The Shah went into exile with billions — but that was an exception. Gaddafi did not leave. Saddam did not leave. Assad left last, and only thanks to a Russian military intervention. For the ruling class of the neo-feudal domains, the loss of power is synonymous with the loss of everything — including physical survival.

Second, feudal empires were not stable among themselves. The standard argument for feudal stability — medieval feudal systems lasted for centuries — ignores a key detail: internally they were stable, but among themselves they waged permanent wars. The Great Wall of China, the Hundred Years' War, the Thirty Years' War, the Napoleonic Wars — these are all conflicts between feudal and proto-national empires. Neo-feudal domains will be stable only if there are no strong imbalances among them — and in a globalized, communication-connected world they will not be able to isolate their subjects as perfectly as was the case in the past.

This is probably one of the reasons for Russian aggression against Ukraine. Putin's regime would be threatened if Ukrainians were to fare significantly better than Russians. In the era of the internet and tourism it is not possible to prevent subjects from seeing that life is better elsewhere. That is why Putin declared that "for us this is a matter of life and death" — not because Ukraine militarily threatened Russia, but because a successful, prosperous Ukraine in the EU would demonstrate the existence of an alternative.

Third, economic interdependence is a weapon, not an insurance policy. China holds a significant portion of American government debt and is actively selling it. The United States is seeking to reduce the value of the dollar. In classical economic theory, mutual interdependence is a guarantee of peace — no one will go to war with a trading partner. But Allison himself concedes the paradox: "The U.S. and China have economies so intertwined that in the event of war, Americans would have nothing to buy in Walmarts and Chinese factories would produce goods that no one buys." Yet it is precisely this interdependence that can become an instrument of economic war before kinetic war — and the mutual decoupling that both sides are actively carrying out reduces the cost of a future conflict.

In 1991 the Soviet Union collapsed. It had thousands of nuclear warheads — and did not use them. This fact is the strongest argument against the thesis of nuclear escalation in the collapse of a neo-feudal empire. But it is an argument that requires closer examination.

The USSR in 1991 collapsed from within. Economic exhaustion after the war in Afghanistan, Gorbachev's reforms that spun out of control, national movements in the Soviet republics, and the structural inefficiency of the planned economy — all this led to a situation in which there was no one against whom to use the nuclear weapons. No one was attacking the Soviet Union. The generals at the button were calculating their personal future in the post-Soviet world. There was no external aggressor against whom a retaliatory launch would make sense.

Both Putin and the Chinese president Xi Jinping learned from 1991. And documented changes in Russian nuclear doctrine reflect this.

In November 2024 Russia formally approved a revised nuclear doctrine. The key change: the previous version (2020) permitted a nuclear response only when "the very existence of the state is threatened." The new version expands the conditions to "a critical threat to sovereignty and/or territorial integrity." The Carnegie Endowment for International Peace analyzes: this formulation allows the regime to conflate threats to its political survival with threats to state sovereignty, thereby effectively using nuclear deterrence to protect internal stability.

CSIS confirms: Putin is lowering the threshold for the potential use of nuclear weapons while simultaneously increasing the ambiguity about when they would be used. He is signaling a willingness to take greater risks. Mark B. Schneider of NIPP (Information Series No. 615, February 2025) goes even further: the language of the document suggests that the doctrine's main objective is a war with NATO within a decade, for which the Russian leadership is, by its own statements, preparing.

The Russian strategic hawk Sergei Karaganov in 2024 directly called for "a rapid ascent up the ladder of nuclear escalation." Putin at the time explicitly rejected the revised escalation — "I do not think such a situation has arisen" — but since then the doctrine has shifted precisely in the direction that Karaganov demanded.

A CSIS study identified, over the course of the war in Ukraine, more than 200 cases in which Russian officials mentioned nuclear weapons in the context of the conflict. Roy Allison, in the article "Averting acute escalation in Russia's war against Ukraine" (International Affairs, Vol. 101, No. 5, September 2025), analyzes how the frequency and directness of Russian nuclear threats has no parallel since the Khrushchev era.

And that is only one neo-kingdom. North Korea has possessed nuclear weapons for 20 years. Pakistan and India face off in permanent tension. Iran is moving toward nuclear capacity. Saudi Arabia is signaling interest. Every new nuclear domain increases the probability that one of them will, in some crisis, cross the threshold — not necessarily by a rational decision, but through a miscalculation, a technical failure, or the desperation of a losing regime.

And it is precisely here that artificial intelligence enters the stage — not as a solution, but as a catalyst.

Artificial intelligence is the first technology in history that has the potential to act simultaneously as a leveler of inequality and as a destabilizer of the nuclear balance. This ambivalence is not a rhetorical figure — it is empirically documented in the academic literature, which since 2024 has exploded along both vectors.

A key IMF study (Rockall, Mendes Tavares & Pizzinelli, Working Paper 2025/068, DOI: 10.5089/9798229006828.001) brings a surprising finding: unlike previous waves of automation, which increased both wage and wealth inequality, AI could reduce wage inequality. The reason is simple: whereas industrial robots replaced routine manual labor (low-paid workers), AI replaces cognitive non-routine work — precisely the work performed by highly paid specialists. Lawyers, radiologists, financial analysts, programmers — these are the professions that AI is transforming the fastest.

David Autor of MIT (NBER Working Paper 32140, 2024) argues even more optimistically: AI can help restore the middle class by enabling less-skilled individuals to perform more complex tasks. A nurse assisted by AI diagnostics approaches the productivity of a physician. A technician with an AI analytical tool approaches the engineer. AI democratizes expertise.

CEPR (Bloom, Prettner, Saadaoui & Veruete, NBER Working Paper 32430, 2024) quantifies it: because AI substitutes predominantly highly skilled labor, it should reduce the skill premium — the difference between the wages of skilled and unskilled workers. This is the exact opposite of what industrial automation did.

If this vector were to dominate, AI could thaw the Kondratiev cycle: a technology that spreads across the whole of society and creates the "golden age" of broad prosperity that the IT revolution failed to deliver. A fifth horseman that levels inequality without violence.

But the same IMF study contains a key caveat: although AI may reduce wage inequality, it will probably substantially increase wealth inequality. The highly paid workers whose tasks AI replaces are at the same time those who own capital — stocks, real estate, investments. The growing returns from AI capital flow to them. The falling wages are compensated by rising capital income. The result: the wage gap narrows, the wealth gap deepens.

Bell and Korinek (Brookings, Journal of Democracy, 2024) warn of a feedback loop: high inequality undermines democratic institutions through the increased influence of elites, corruption, and populism. At the same time, a weakened democracy loses the ability to rein in inequality through progressive policies. AI-driven inequality could accelerate this vicious circle.

And the global dimension is even more alarming. The Center for Global Development documents the extreme asymmetry of AI investment — as stated above, in 2023 the United States received 8.7 times more private AI investment than China, the second in the ranking. China is the only country below the high-income threshold among the 30 most innovative countries. Internet access is 27% in low-income countries versus 93% in wealthy countries. AI threatens that the deepening of the technological gap between countries will reverse the decline in inter-state inequality that was the main positive story of globalization since 2000.

But the truly existential risk of AI lies elsewhere — in nuclear strategy.

Mutually Assured Destruction (MAD) — mutual guaranteed destruction — is the doctrine that has prevented nuclear war since 1945. Its logic is simple: no state will launch a nuclear attack, because it knows that the adversary will respond with a retaliatory strike that destroys the attacker. The key to MAD is the capability for a retaliatory strike — the certainty that even after absorbing a first strike, enough nuclear forces will remain for a devastating response. Hence ballistic-missile submarines (SSBNs), mobile missile launchers, and dispersed command structures.

AI systematically erodes this balance.

Lieutenant General John Shanahan, the former director of the Pentagon's Joint Artificial Intelligence Center, published in September 2025 in Arms Control Association a cautionary analysis: a particularly destabilizing scenario involves the large-scale use of AI to detect and continuously monitor an adversary's nuclear forces. The integration of AI systems capable of reliably locating mobile missile launchers or submerged ballistic-missile submarines — considered the most resilient elements of retaliatory capability — could undermine confidence in the capacity for a retaliatory strike. The loss of this confidence increases the incentive for a preemptive strike.

The RAND Corporation, in an extensive study (How Might Artificial Intelligence Affect the Risk of Nuclear War?), elaborated the mechanism: conventional weapons with AI-assisted targeting could be perceived as capable of tracking and destroying enemy launchers. An adversary threatened with the loss of its retaliatory-strike capability would be pushed toward a preemptive first strike or toward expanding its arsenal.

SIPRI (Stockholm International Peace Research Institute, June 2025) adds another dimension: the opaque recommendations of an AI decision-support system can tilt decision-makers toward action. A human in the decision chain who is told by an AI system "attack detected with 94% probability" will react differently from a human analyzing raw data.

And then there is the compression of decision times. During the Cuban Missile Crisis in 1962, Kennedy and Khrushchev had thirteen days to negotiate. In the era of hypersonic missiles and AI-assisted targeting, the decision window compresses to minutes. Chatham House (September 2025) warns: the key risk of neural networks in nuclear systems is that the AI will make a wrong decision and the human operator will not have enough time or information to challenge it.

Fascinating — and disturbing — is a recent experiment. A preprint from February 2025 (arxiv.org) tested the most advanced AI models in simulated nuclear crises according to Herman Kahn's escalation ladder. The results: Google Gemini was the only model that deliberately chose strategic nuclear war — it did so in a first-strike scenario by the fourth round — and the only model that explicitly invoked the "rationality of irrationality." The models matched their own signals roughly 70% of the time, but with dramatic differences: Claude and GPT-5.2 achieved 72–75% consistency, while Gemini only 50%, reflecting its strategy of deliberate unpredictability.

These are not science-fiction scenarios. General Anthony Cotton, commander of U.S. Strategic Command, testified before the Senate in March 2025: "USSTRATCOM will use AI/ML to enable and accelerate decision-making." Biden and Xi Jinping agreed in November 2024 that humans, not AI, should decide on the use of nuclear weapons — but the agreement is non-binding, and the Trump administration rescinded Biden's executive order on AI.

And here the argument about the neo-feudal domains returns. AI does not act on the nuclear balance symmetrically. The technologically dominant domains — the United States and China — gain AI-assisted surveillance capabilities that gradually erode the retaliatory-strike capability of the lagging domains. Russia, whose conventional forces are weakened by the war in Ukraine, is increasingly dependent on its nuclear arsenal as the last guarantee of relevance. If AI-driven surveillance one day makes it possible to locate Russian submarines and mobile launchers in real time, Russia will lose its retaliatory-strike capability — and with it the only insurance for the regime's survival.

It is precisely in this scenario — a losing neo-feudal domain facing the loss of its last guarantee — that it becomes rational to use nuclear weapons before losing them. "Use it or lose it" is not a paranoid fantasy — it is a doctrinal scenario that Russian strategic planners are actively modeling.

And this applies not only to Russia. Every neo-feudal domain with a nuclear arsenal — India, Pakistan, North Korea, potentially Iran — faces a similar calculative pressure if AI-driven surveillance by one domain threatens the retaliatory capabilities of another.

The situation at which we have analytically arrived is this: the world is heading neither toward an unambiguous AI utopia nor toward an inevitable nuclear apocalypse. It is heading toward a race — between two vectors of the same technology.

The democratizing vector needs: the spread of AI tools into education, healthcare, and the productivity of broad strata; regulation that prevents the monopolization of AI benefits; the redistribution of profits from AI productivity; decades of building infrastructure in developing countries. Autor, Bloom, and others show that the mechanism exists — AI can reduce the skill premium and restore the middle class. But its realization requires political will, international cooperation, and time.

The militarizing vector needs: continued investment in AI-supported surveillance, autonomous weapons, and nuclear decision systems; the absence of international agreements on the regulation of military AI; continued geopolitical rivalry. The Pentagon and the PLA are making these investments right now. The feedback loop is short: investment → capability → demonstration → adversary's reaction → escalation.

The race is unfair. The democratizing vector has a delayed feedback loop — its effects manifest over a horizon of decades. The militarizing vector has a feedback loop measured in years. Surveillance AI for submarines is funded by state budgets today. Democratizing AI for education in Africa needs infrastructure that no one has yet built there.

But the race is not hopeless. There are three factors that could accelerate the democratizing vector:

First, AI is by its nature diffusive. Unlike nuclear weapons, where a technological monopoly lasts decades, AI models spread rapidly. Open-source models (Llama, Mistral, Qwen) are available to anyone with a GPU. This does not mean equality — access to computing capacity is extremely unequal — but it does mean that a knowledge monopoly is difficult to sustain.

Second, the economic incentive. McKinsey estimates the global economic impact of AI at 4.4 trillion dollars per year by 2030. A significant part of this impact flows from applications that expand productivity — not merely concentrate it. Firms that make AI accessible to a broader workforce will have a competitive advantage over those that use it only to replace workers.

Third, the historical precedent. Every previous universal technology — the steam engine, electricity, the internet — passed through a phase of concentration followed by diffusion. Perez's model predicts that after the phase of speculation and crisis comes a "golden age" in which the technology spreads across the whole of society. The question is not whether, but when — and whether this can be accomplished before the militarizing vector reaches a critical point.

The analytical chain presented in this article — from Turchin's cycles through the neutralization of Scheidel's horsemen to the neo-feudal domains and the race of the two AI vectors — is coherent. But coherence is not the same as truth. Fair treatment of the criticism requires naming the weak links.

First, the falsifiability of cyclical theories. Yascha Mounk of Johns Hopkins calls Turchin's theory unfalsifiable — the predictions are so general that practically any turbulent event will confirm them. Georgescu (2023, PLoS ONE), in formal testing, found that the model's predictions "are not supported by the data." Kondratiev waves are disputed — the periodization differs from author to author. With 210 theories of Rome's fall, as the historian Alexander Demandt counted, one can retrospectively fit the data to many frameworks.

Second, eighty years without a nuclear war. Since 1945 humanity has possessed nuclear weapons — and has never used them in war. North Korea has had them for 20 years. India and Pakistan have faced off for decades. The USSR collapsed with thousands of warheads and did not use them. Schelling (2005, Nobel): nuclear deterrence is the "most successful public policy" of the modern era. The argument that MAD will be eroded is logical, but empirically unsubstantiated — to date no technology has broken MAD.

Third, economic interdependence. Allison himself acknowledges that the United States and China are economically intertwined to an unprecedented degree. Feudal lords did not clash when mutual trade was more profitable. Global supply chains create mutual dependencies that raise the cost of conflict far above the level of any historical precedent. Liberal theory of international relations (Keohane, Nye) argues that it is precisely this interdependence that is the strongest peace mechanism.

Fourth, feudal systems can be stable. Medieval feudalism lasted for centuries. Caste India for millennia. If the neo-feudal domains consolidate power, they have a strong incentive not to go to war — war threatens the structure guaranteeing privilege. Varoufakis himself argues that the main U.S.–China rivalry is a rivalry of cloud domains, not necessarily a military one.

Fifth, AI optimism is not naive. The Scandinavian model — the Saltsjöbaden Agreement, wage compression — demonstrably reduced inequality before World War II, without violence. The New Deal responded to the crisis with reforms, not revolution. There is no a priori reason why the democratizing vector of AI could not trigger an analogous reform wave — if sufficient political pressure exists.

What remains valid even after taking these objections into account: The data on inequality are hard and independent of any cyclical theory. A Gini index of 0.49. The top 1% share at 21%. A CEO-to-employee pay ratio of 285:1. The neutralization of Scheidel's horsemen is empirically documented — not speculative. The lowering of the Russian nuclear threshold is a documented fact, not an interpretation. And AI really does act in both vectors simultaneously — this is not a thesis, but a description of reality confirmed by the IMF study as well as by analyses from RAND, SIPRI, and the Arms Control Association.

We do not claim that nuclear war is inevitable. We claim that the historical corrective mechanisms of inequality have been taken out of service, that the trajectory toward neo-feudal domains is empirically grounded, and that AI is the only remaining variable that can change the trajectory — in both directions.

Let us return to the number with which we began. The Gini coefficient of American households: 0.49 in 1928 and 0.49 in 2024. Back then, corrective mechanisms existed — cruel, bloody, but functional. Today they have been taken out of service. Nuclear weapons transformed great war into an existential threat. Conventional wars deepen inequality. Pandemics in the era of modern medicine and quantitative easing paradoxically enrich the rich. Revolution and collapse run up against the digital surveillance and predictive repression of the modern state.

What remains are two minute hands on a single dial. One counts down to AI democratization — to a world in which artificial intelligence spreads expertise into broad strata, reduces wage inequality, restores the middle class, and thaws the Kondratiev cycle frozen since 2008. To a world where a fifth horseman levels inequality without violence.

The other counts down to AI destabilization — to a world in which artificial intelligence erodes mutually assured destruction, compresses decision times from days to minutes, asymmetrically favors technologically dominant powers, and pushes lagging nuclear domains toward the logic of "use it or lose it." To a world where the very AI that was supposed to be the solution becomes the catalyst of catastrophe.

Which hand will reach the end first?

To this we do not yet have an answer. But unlike in 1928, we know that the countdown is running — and we know what is at stake. For the first time in history we have analytical tools, data, and models that allow us to see trajectories before they are fulfilled. Turchin's cycles, Scheidel's compressions, Piketty's r > g, Allison's Thucydides Trap, Varoufakis's technofeudalism — none of these are prophecies. They are diagnostic tools.

And the diagnosis says that treatment is still possible. But the window is closing.

The article is based on a synthesis of academic literature from the fields of cliodynamics, the economics of inequality, nuclear strategy, and artificial-intelligence research. Key primary sources: Turchin (2016, Ages of Discord; 2023, End Times), Scheidel (2017, The Great Leveler), Piketty (2014, Capital in the Twenty-First Century), Piketty & Saez (2003, QJE), Goldin & Margo (1992, QJE), Scheve & Stasavage (2016, Taxing the Rich), Acemoglu & Robinson (2012, Why Nations Fail), Perez (2002, Technological Revolutions and Financial Capital), Milanovic (2016, Global Inequality), Zucman (2015, The Hidden Wealth of Nations), Hauner, Milanovic & Naidu (2020, Stone Center WP), McCoy, Roscoe et al. (2025, PNAS, DOI: 10.1073/pnas.2400695121), Heldring (2023, EEH), Peltier (2025, Dædalus), Stiglitz & Bilmes (2008, The Three Trillion Dollar War), Varoufakis (2024, Technofeudalism), Allison (2017, Destined for War), Skocpol (1979, States and Social Revolutions), Guriev & Treisman (2022, Spin Dictators (Informační diktátoři)), Feldstein (2021, Carnegie), Xu Xu (AJPS), Schneider (2025, NIPP Information Series No. 615), Russian nuclear doctrine (2024, document No. 991), CSIS analysis (January 2025), Carnegie Endowment (November 2024), RAND (How Might AI Affect Nuclear Risk), Shanahan (2025, Arms Control Association), SIPRI (June 2025), IMF Working Paper 2025/068 (Rockall et al., DOI: 10.5089/9798229006828.001), Autor (2024, NBER WP 32140), Bloom et al. (2024, NBER WP 32430), Bell & Korinek (2024, Brookings/Journal of Democracy), CGD (2024), AI nuclear crisis simulation (arxiv 2602.14740, February 2025), Roy Allison (2025, International Affairs, Vol. 101/5), World Inequality Database (WID.world), World Inequality Report 2022, Oxfam Inequality Reports, AFL-CIO Executive Paywatch (2025), the Seshat/CrisisDB database.

Transparency of creation:

The conception, structure, and editorial line of the article are the work of the author, who prepared the content sketch, established the key theses, and directed the entire creation process. Generative AI (Claude, Anthropic) was used as a technical tool for research, fact-checking, and elaborating the author's outline.

The author edited the outputs throughout, verified the key findings, and approved the final wording. No part of the text was published without human review. All factual data were verified against the publicly available sources cited in the text.

The procedure complies with the requirements of Article 50 of EU Regulation 2024/1689 (AI Act) on the transparency of AI-generated content. #poweredByAI