AI Great Divergence | Economic Impact & Policy

The AI Great Divergence — A New Economic Era

The Council of Economic Advisers’ January 2026 report frames artificial intelligence through the lens of economic history’s most consequential transformation: the Industrial Revolution’s “Great Divergence.” Between 1800 and 1900, industrializing nations pulled dramatically ahead of the rest of the world economically, creating persistent global inequalities that shaped centuries of development. The report poses a critical question: if AI proves as transformative as steam power and mechanization, will it create a second Great Divergence between nations that lead in AI development and those that lag behind?

This framing carries profound implications because recent economic history has actually witnessed a “great convergence” — the last 25 years saw developing nations growing faster than rich ones, narrowing global income gaps and lifting billions from poverty. China’s rise, India’s software revolution, and broad-based emerging market growth characterized this era of economic catch-up. But the Council warns that AI could dramatically reverse this trend. Nations that achieve AI leadership may experience economic acceleration so pronounced that it recreates the divergent growth patterns of the Industrial Revolution, potentially leaving AI laggards further behind than ever.

The report establishes clear definitions for understanding AI’s current state and trajectory. It distinguishes between narrow AI — systems that may be superhuman at specific tasks but cannot perform all human intellectual work — and the hypothetical future possibilities of Artificial General Intelligence (AGI) and Artificial Superintelligence (ASI). Crucially, major AI developers including OpenAI, Anthropic, xAI, Meta, and Google all explicitly aim to create AGI or superintelligence, though the report notes that AGI implications are “outside the scope” of current analysis, focusing instead on the economic impacts of narrow AI systems that exist today and are rapidly improving. For deeper analysis of how AI is reshaping global economic power, explore our interactive guide to AI and global economic competition.

AI’s GDP Impact — From Conservative to Revolutionary Estimates

The economic literature on AI’s GDP impact reveals a striking range of predictions that span from modest productivity gains to transformational economic acceleration. Conservative estimates include Acemoglu’s projection of 0.9-1.6% GDP increase over 10 years and Penn Wharton’s 1.5% estimate over the same period. Mid-range projections include McKinsey’s global estimate of 2.4-4.1% long-run impact and Goldman Sachs’ prediction of 7% global GDP growth over 10 years. The most aggressive estimates reach truly revolutionary territory: PwC projects 1-15% growth over 10 years, Aldasoro et al. estimate 20-45% over 10 years, and Robin Hanson’s theoretical AGI scenario suggests growth rates exceeding 45% annually.

What makes these projections particularly compelling is that the lowest estimates are already being challenged by actual investment data. AI-related investment increased US GDP by an annualized rate of 1.3% in the first half of 2025 alone — a figure the Council notes is comparable to railroad investment’s contribution during the Industrial Revolution. This single data point suggests that even conservative GDP impact estimates may prove too low if AI investment continues at current rates. For comparison, a 2010 ITIF study found the entire IT revolution boosted US GDP by approximately 14% over several decades, providing context for what transformative technology adoption can achieve.

The wide range of estimates reflects genuine uncertainty about AI’s trajectory, but also highlights the potential for exponential rather than linear economic effects. The report notes that information processing equipment and software investment grew at a 28% annual rate in H1 2025, up from 5.5% in 2024, with Q2 2025 showing investment already $125 billion higher (in annual terms) than end-of-2024 levels. This category now represents 25% of all US investment, a proportion that suggests AI’s economic footprint is expanding rapidly beyond technology companies into the broader economy.

Total Factor Productivity as the Growth Engine

For rich countries with high capital stocks like the United States, the Council emphasizes that sustained economic growth primarily comes from increasing Total Factor Productivity (TFP) — the portion of economic output not explained by increases in capital or labor inputs. TFP captures the efficiency gains from innovation, better organization, and technological progress, making it the most important long-term determinant of living standards in developed economies. The report positions AI as potentially the most significant TFP driver since the Industrial Revolution, with implications for economic growth that could persist for decades.

However, TFP is fundamentally a lagging indicator — the productivity effects of new technologies emerge with significant time delays as businesses adopt, adapt, and optimize their use of innovations. The report provides historical context: productivity gains from 1990s technology investments largely emerged from tech spending in the 1970s and 1980s, while investments made during the Great Depression bore fruit in the economic boom of the 1950s and 1960s. This pattern means that current AI investment may not show up in productivity statistics for years, echoing Robert Solow’s famous 1987 observation that “you can see the computer age everywhere but in the productivity statistics.”

Given TFP’s lagging nature, the Council advocates tracking leading indicators of future productivity growth: R&D spending on AI, AI firm output growth, and the performance metrics of AI systems themselves. Current data on these leading indicators is striking: R&D spending as a percentage of GDP varies dramatically across countries, with Israel at 6.0%, South Korea at 5.2%, and the US at 3.6%, compared to just 2.1% for the EU. Meanwhile, AI firm revenues are growing exponentially — OpenAI’s revenue jumped from approximately $350 million in H1 2023 to $2.4 billion in H2 2024, with the company projecting roughly doubled revenue each year through 2028.

Jevons’ Paradox and the Future of Work

The report addresses widespread concerns about AI-driven unemployment through the lens of Jevons’ Paradox, a counterintuitive economic principle where efficiency improvements actually increase total resource consumption rather than decrease it. Named after economist William Stanley Jevons, who observed in 1865 that improved coal efficiency led to increased total coal demand, the paradox suggests that AI productivity gains could actually increase employment rather than reduce it — if certain conditions are met.

For Jevons’ Paradox to apply to AI and labor, three conditions must hold: AI must meaningfully boost worker productivity, cost savings must translate to lower prices for goods and services, and lower prices must increase consumer demand faster than efficiency gains reduce per-unit labor requirements. Historical precedent strongly supports this possibility. Examples of Jevons’ Paradox include irrigation efficiency leading to increased water consumption, energy-efficient lighting increasing electricity use for lighting applications, road capacity increases generating more drivers, and counterintuitively, radiologists — once predicted to be replaced by AI — now experiencing historically high employment levels.

Current labor market evidence presents a mixed picture that neither confirms nor refutes AI displacement fears. Research by Brynjolfsson et al. finds employment falling for early-career workers in AI-exposed occupations like coding and customer service, while other studies show no correlation between AI exposure and unemployment rates across different sectors. More encouragingly, Johnston and Makridis found that employment fell in sectors where AI directly substitutes for labor but increased in sectors reliant on AI-capable tasks, suggesting a rebalancing rather than wholesale displacement. The historical precedent is also reassuring: in 1860, 43% of US employment was in agriculture compared to 1.2% in 2015, yet the majority of current workers are in jobs that didn’t exist in 1940 — from wind turbine technicians to software developers to mental health counselors.

The Speed of Change Defines the AI Revolution

The Council identifies the speed of AI advancement as perhaps the most defining characteristic of the current technological transformation, with metrics improving at rates that dwarf historical technology adoption curves. Training compute for AI systems has increased by approximately 4x per year since 2010, representing over a billion-fold increase since 2012. Top AI company revenues are tripling annually, many performance metrics are doubling every few months, and the length of tasks AI can complete is doubling every 7 months — a trend that has persisted for six years.

The cost improvements are equally dramatic. Token prices — the units in which AI language models process text — are falling at rates between 9x and 900x per year, making AI applications economically viable for use cases that were prohibitively expensive just months earlier. Performance improvements match the cost declines: AI performance on SWE-bench, a coding evaluation benchmark, improved from 4% to 72% in just one year between 2023 and 2024. These rapid improvements suggest that “the AI of the future will likely be very different from the AI of today,” as the report notes, making long-term predictions both crucial and challenging.

This acceleration has profound implications for economic planning and policy formation. Traditional technology adoption curves suggest that transformative technologies take decades to achieve broad economic impact, but AI’s rapid improvement trajectory compresses these timelines dramatically. The report notes that current AI agents still struggle with longer task sequences, meaning they cannot yet fully substitute for even low-skill computer-based work, but this capability is improving on a doubling trajectory that could overcome current limitations within years rather than decades. The Bureau of Labor Statistics employment data provides ongoing tracking of these labor market dynamics. The speed of change demands that policymakers, businesses, and individuals adapt their planning horizons and preparation strategies accordingly. Our AI workforce transformation guide explores how organizations can prepare for these rapid changes.

US AI Dominance Across Investment and Infrastructure

The United States has established commanding leadership across most measurable AI metrics, creating the foundation for disproportionate economic benefits from AI advancement. America controls 74% of the world’s compute capacity for AI training and inference as of May 2025, a concentration that represents unprecedented technological dominance in a field that increasingly drives economic growth. US private AI investment reached $109 billion in 2024, more than ten times China’s $9 billion and representing approximately 75% of reported global venture funding in generative AI startups.

The investment dominance extends beyond single-year figures to cumulative advantage: the US has attracted over $470 billion in cumulative private AI investment between 2013 and 2024, compared to approximately $50 billion for all EU countries combined. This 10-to-1 advantage in cumulative investment translates directly into infrastructure, talent, and technological capabilities that compound over time. The US also leads in large-scale AI system development, with 154 such systems compared to China’s significant but smaller numbers, and hosts the majority of leading AI research labs and companies.

However, the report notes important nuances in American AI leadership. While the US dominates investment and infrastructure, the performance gap between frontier AI models from different countries is relatively narrow — just 11 months separate the most advanced models from different nations. This suggests that while building AI infrastructure requires enormous capital and resources that favor wealthy nations, the knowledge and techniques for developing competitive AI models can spread more quickly. The tension between infrastructure advantages and knowledge diffusion will likely shape how AI economic benefits distribute globally in the coming decade. To understand the broader implications of this technological leadership, see our interactive analysis of US AI global leadership strategies.

International Competition — Winners and Laggards

The global AI competition reveals clear winners, emerging challengers, and concerning laggards that illuminate how the AI Great Divergence may unfold. China ranks second across most AI metrics but remains heavily dependent on US technology infrastructure — almost all Chinese AI models are trained on US-manufactured hardware, creating strategic vulnerabilities that could limit China’s AI development if technology export restrictions tighten. China’s large public sector AI spending, approximately $56 billion in 2025, demonstrates government commitment but cannot fully compensate for private sector investment gaps or supply chain dependencies.

The European Union presents perhaps the most concerning case of AI economic divergence. The EU’s share of world GDP has fallen from 27% in 1980 to just 14% in 2025, and the region significantly lags in AI investment, infrastructure, and commercial deployment. Germany and many other EU countries have growth rates lower than other advanced economies, suggesting that pre-existing economic challenges may be compounded by slower AI adoption. The EU’s $50 billion cumulative AI investment pales compared to US levels, and regulatory approaches that prioritize safety and privacy over innovation may further slow adoption and economic benefits.

Emerging challengers include Middle Eastern nations making substantial AI investments with sovereign wealth fund resources. Saudi Arabia’s Public Investment Fund established Humain with a $10 billion venture fund specifically for AI, while the UAE is partnering with OpenAI and NVIDIA on the Stargate project for AI infrastructure development. These nations leverage oil wealth to purchase AI capabilities and infrastructure, potentially allowing them to leapfrog traditional economic development stages. The report introduces the concept of “Pax Silica” — America’s international AI supply chain partnership — noting that member countries are growing more than twice as fast as peers, with 2.5% average real GDP growth compared to 1.1% for G7 countries.

Trump Administration AI Policy Framework

The Trump administration’s AI strategy centers on three integrated policy pillars designed to maintain American AI leadership while maximizing economic benefits. The centerpiece legislative achievement is the “One Big Beautiful Bill Act,” signed July 4, 2025, which restored 100% bonus depreciation for IT infrastructure and data center equipment. The Council predicts this single policy change will generate over 1% GDP growth per year for four years, increase real wages by $4,000-$7,200 per worker, and boost investment levels by 7-10%. This demonstrates how targeted tax policy can accelerate private sector AI adoption and infrastructure development.

The AI Action Plan, launched in July 2025, establishes three strategic priorities: rapid data center buildout, enabling innovation through reduced regulatory barriers, and upholding free speech principles in AI model development. Specific implementing actions include NEPA categorical exclusions to streamline data center permitting, establishment of AI Centers of Excellence across federal agencies, and federal procurement guidelines requiring objective and unbiased AI systems. The deregulation component includes executive orders signed July 23 and December 11 to accelerate data center permitting and reduce state-level barriers, with the Council estimating these measures alone could add 0.3-0.8 percentage points to annual GDP growth for two decades.

The energy dominance initiative recognizes that AI data centers require unprecedented amounts of electricity, making energy policy inseparable from AI strategy. The administration resumed federal energy leasing, approved new LNG export permits, and increased support for advanced nuclear technologies. The Council estimates these energy policies could increase GDP by 0.3-1.2% by 2035 while reducing costs for AI infrastructure development. The Department of Energy identified 16 federal sites for rapid data center construction in April 2025, while supporting grid management technologies and power line upgrades to accommodate AI energy demands. This comprehensive approach demonstrates recognition that AI leadership requires coordinated policy across multiple domains rather than technology policy in isolation.

Energy Infrastructure and AI Data Center Demands

AI’s massive energy requirements represent both a critical bottleneck and a significant economic opportunity that could reshape global energy markets and infrastructure development. The report projects that AI data centers will account for 4% of US electricity demand in 2023, rising to 7-12% by 2028 — a level of consumption that requires substantial new generation capacity and transmission infrastructure. Training advanced AI models like Grok 4 cost approximately $490 million in July 2025, with much of that expense representing electricity consumption for powering massive GPU clusters over months of training time.

The economic implications of AI energy demand extend far beyond electricity costs to encompass entire supply chains and geographic development patterns. Cloud compute costs are growing at 2.5x per year on average, driving demand not just for electricity but for reliable, low-latency power delivery that can support always-on AI operations. This creates opportunities for regions and countries with abundant, cheap, and reliable energy sources to become AI infrastructure hubs, while potentially disadvantaging areas with expensive or unreliable electricity supplies.

The administration’s energy dominance strategy directly addresses these requirements through multiple channels. Resumed federal energy leasing expands domestic oil and gas production, new LNG export permits support AI-powered economic growth through energy exports, and advanced nuclear support targets the reliable baseload power that AI data centers require. The Council’s economic modeling suggests these policies could boost GDP by 0.3-1.2% by 2035, demonstrating how energy abundance translates directly into AI economic advantages. Countries that cannot provide cheap, reliable electricity for AI data centers may find themselves excluded from the most economically valuable AI applications, creating another dimension of the potential Great Divergence.

Tracking AI Progress — Key Metrics and Indicators

The Council recommends monitoring specific metrics across three categories to track AI economic development and predict future outcomes: investment flows, performance improvements, and adoption rates. Investment metrics include both private venture capital and corporate AI spending, AI model training costs, and infrastructure investments in data centers and specialized computing hardware. These leading indicators provide early signals of where AI capabilities and economic benefits are likely to emerge, similar to how railroad investment predicted Industrial Revolution economic impacts.

Performance metrics capture the rapid capability improvements that drive AI economic potential. Key indicators include benchmark score improvements across tasks like coding (SWE-bench), reasoning, and domain-specific applications, falling cost per token or computational operation, and the increasing length and complexity of tasks AI systems can complete autonomously. The report emphasizes tracking the doubling times of these metrics — currently ranging from every few months to every seven months for task length — as indicators of how quickly AI economic applications may expand.

Adoption and usage metrics measure how quickly AI capabilities translate into real economic activity. Critical indicators include the percentage of organizations using AI in production (rising from under 4% to approximately 10% for US firms), business revenue from AI applications, the share of workers using AI tools (currently around 40% for US knowledge workers), and consumption of critical minerals required for AI hardware manufacturing. The report notes that ChatGPT usage shows US represents only 19% of traffic despite American AI leadership, suggesting global demand for AI capabilities may create export opportunities even for countries that cannot develop frontier AI systems domestically.