Current thematic dossiers

Select a theme to view its dossier.

Artificial intelligence

This deep dive examines the latest developments in artificial intelligence, including emerging adoption patterns, key risk watch points, evolving policy and supervisory guidance, and research on the economic impact of AI.

Overview

The economic impact of AI is becoming more visible, but remains uneven and highly conditional. Current evidence shows measurable productivity gains in specific tasks and among adopting firms, while the translation into sustained economy-wide growth depends on the pace of diffusion and complementary investments in infrastructure, data, software, skills and organisational change. Labour market effects have so far been selective rather than systemic, with limited evidence of broad job displacement but signs of pressure in hiring for AI-exposed and entry-level roles. The most pronounced macroeconomic effect to date is on investment. Capital expenditure on data centres, semiconductors and related infrastructure has increased substantially, is increasingly financed through debt and capital markets, and is lifting high technology imports and near-term price pressures ahead of any supply-side gains. At the same time, AI production is highly concentrated across a small number of firms, jurisdictions and supply chain nodes, creating both innovation benefits and strategic dependencies. This deep dive examines these dynamics through four lenses: productivity and labour market adjustment, AI readiness, the structure and concentration of AI production, and the AI investment boom and its macroeconomic effects.

What's new

Federal Reserve Board Chairman Kevin Warsh outlined staff scenario analysis of super intelligence in opening remarks at the Productivity and Jobs Task Force Conference. The exercises consider sustained productivity gains with full employment, stronger growth accompanied by job displacement, and a slowdown following disappointing investment returns. They are illustrative scenarios rather than forecasts, intended to support preparation for potential monetary policy implications.

The Dutch central bank (DNB) has published research finding that smaller firms lag larger companies in artificial intelligence adoption and that productivity benefits require integration into everyday operations. Employees use AI more often and save more time when employers provide active support, including training and access to tools. Cloud services, standard software and practical implementation advice can help smaller businesses overcome resource and knowledge constraints.

Speaking at Yale University, Federal Reserve Board Governor Michael S. Barr warned that AI could worsen inequality if dominant firms and capital owners capture most of its returns. Although worker displacement remains limited and competition has kept access to AI fairly broad, rapid adoption could cause substantial labor market disruption. He urged early consideration of job creation, worker training and support for transitions to new roles.

Deep dive

AI productivity and labour market

AI is improving productivity in some tasks and firms, while labour-market effects are so far concentrated in particular occupations and entry-level roles.

Research points to measurable productivity gains in individual tasks and among firms that have adopted AI. The translation of these gains into economy-wide productivity growth remains uncertain and appears to depend on how widely and how intensively AI is adopted and whether firms also invest in software, data, training and organisational change. Employment data do not yet show broad-based job losses or a general decline in hiring. However, weaker outcomes are emerging in AI-exposed occupations and entry-level roles, appearing first in reduced hiring of young workers rather than in total employment, particularly where AI can replace rather than support workers’ tasks. Demand for new skills is associated with higher wages and, in some cases, stronger employment, but AI-specific skills have so far been linked mainly to wage premia rather than overall job growth. Taken together, the current evidence points to selective labour-market adjustment rather than widespread displacement.

  • The Bank for International Settlements reports productivity gains of 10–65% for selected tasks, compared with estimates of economy-wide annual total factor productivity gains ranging from 0.07% to 0.3–0.9 percentage points.
  • A Bank for International Settlements working paper using European Investment Bank data on more than 12,000 European Union firms estimates that AI adoption raises labour productivity by around 4%, with no adverse short-term employment effect and larger gains among firms that also invest in software, data or employee training.
  • The Board of Governors of the Federal Reserve System found no reduction in job postings among firms or industries with higher AI adoption, while the Bank of Canada reports that almost 90% of Canadian businesses using AI made no change to staffing with only 4% having reported the creation of jobs and 6% the reduction of employment.
  • The International Monetary Fund finds that, in occupations where AI is more likely to replace workers’ tasks than assist them, employment was 3.6% lower in regions with stronger demand for AI skills than in regions with weaker demand, measured five years after those AI skills first appeared in job postings
  • The European Central Bank reports that workplace AI use among euro area workers doubled from 26% in 2024 to 52% in 2026, with the median user saving three hours a week, or 7.7% of median working time, but an estimated economy-wide efficiency gain of only about 3.8% because fewer than half of workers both use AI and report time savings.
  • The Bank of France finds that permanent-contract hiring of workers aged 21 to 29 in the occupations most exposed to generative AI was about 28% lower in March 2026 than in November 2022, against an 8% decline in the least exposed occupations, while total employment showed no significant divergence.
  • Danmarks Nationalbank finds that firms adopting AI in 2023 or early 2024 had employment about 11% below their previous trend by mid-2025 relative to non-adopters, reflecting fewer new hires, particularly of younger and more highly educated workers, with no clear effect on total employment or wages.

AI country readiness

A country’s ability to turn AI adoption into economic gains depends on its digital infrastructure, workforce skills, institutions, investment capacity and economic structure.

Research describes AI readiness as a country’s capacity to adopt the technology at scale and use it productively. This depends on reliable digital infrastructure, a skilled and adaptable workforce, effective institutions and regulation, sufficient investment capacity, and firms that can integrate AI into their operations. Economic structure affects how quickly these capabilities translate into adoption: professional, financial and information-intensive sectors offer more immediate applications for generative AI than agriculture, transport and construction. Advanced economies generally perform better on broad preparedness measures, although readiness varies considerably within both advanced and emerging-market groups. Broad preparedness and AI-specific capacity are not the same. A country may have strong digital and institutional foundations while remaining weaker in AI investment, computing capacity or domestic AI production.

  • The Bank for International Settlements estimates that a standardised improvement in AI preparedness is associated with an average short-run increase in real value-added growth of 0.6 percentage points in advanced economies and 0.45 percentage points in emerging market economies.
  • The International Monetary Fund’s Skill Readiness Index places Finland, Ireland and Denmark among the countries best positioned to equip their workforces with emerging skills relevant to the AI transition, supported by strong tertiary education and lifelong learning.
  • The European Central Bank reports that, among euro-area firms not using AI, 30% cite a lack of perceived usefulness, while around 20% cite incompatibility with existing systems and a similar share cite skill shortages—highlighting firm-level capacity to identify applications, integrate technology and supply relevant skills as distinct dimensions of readiness.
  • The Bank of Canada reports that Canadian business adoption increased from about 3% in 2022 to 12% in 2025, but ranged from 1.5% in accommodation and food services to more than 30% in finance and insurance, illustrating how a country’s sectoral composition affects the scope for near-term AI diffusion.
  • The National Bank of Moldova identifies structured data, digitised processes and accumulated institutional capabilities as prerequisites for extracting economic value from AI.
  • The OECD identifies computing capacity as a strategic economic and security asset needed to develop, scale and access AI systems, finds that only Canada and the United Kingdom have dedicated national compute strategies among 47 jurisdictions surveyed, and calls on governments to assess their domestic compute needs and align investment with them.
  • The International Monetary Fund estimates that AI could raise annual growth in the Middle East and Central Asia by 0.1 to 0.6 percentage points, with gains in Gulf economies about 1.5 times those in the Caucasus and Central Asia and twice those in the region’s low-income economies, a gap it attributes to differences in exposure to AI, preparedness and access to advanced technologies.

Global AI production and market concentration

AI production is concentrated in a small number of firms and economies, giving leading providers growing influence over critical inputs, investment and innovation.

Research describes AI production as a five-layer system comprising computing power, infrastructure, data tools, models and applications. Production is concentrated geographically, with the United States and China hosting most large AI-producing firms and a smaller group of economies occupying important positions in advanced chips and other critical inputs. Most economies specialise in only part of the supply chain, making AI production and deployment dependent on cross-border access to complementary technologies and services. The largest US and Chinese technology firms are expanding across several layers, combining scale in individual markets with a broader presence across the AI ecosystem. Their size and reach give them growing influence over investment, access to key inputs and the direction of innovation. Concentration is most pronounced in computing power, chips and cloud infrastructure, while the model layer has shown greater contestability, with rapidly falling prices and frequent changes in technological leadership. Concentration can support efficiency and rapid technological development, but it also creates dependencies on a limited number of firms, jurisdictions and supply chain nodes.

  • The Bank for International Settlements maps 1,246 AI-producing firms across 32 economies, including nearly 700 in the United States and around 250 in China. Concentration is even greater among the largest firms: the seven largest publicly listed US AI firms were worth more than twice the next 13 global AI firms combined, while US AI giants expanded from an average of about two supply-chain roles in the early 2000s to three or four in 2020–24.
  • The International Monetary Fund identifies concentrated gains in AI-related equities and borrowing by critical unlisted AI firms as channels through which developments in a relatively narrow group of companies could have wider economic and financial effects.
  • The European Central Bank reports that the AI boom has contributed to a surge in global trade in high-technology goods, particularly semiconductors, with the United States acting as a substantial net importer and most of the largest AI supply-chain firms located in the United States, China, Chinese Taipei and South Korea.
  • Danmarks Nationalbank reports that the United States attracted approximately USD 580 billion in venture capital for AI companies during 2021–25, compared with just under USD 76 billion in the European Union—equivalent to annual averages of 0.4% and 0.1% of GDP, respectively.
  • The National Bank of Moldova highlights that concentration in data infrastructure and frontier AI models can create systemic dependencies and widen the divide between countries able to develop AI-based growth opportunities and those dependent on external providers.
  • The OECD finds that the number of developers of language models for cognitive tasks rose from nine in January 2024 to 47 in April 2026 and that quality-adjusted prices for text-to-text models fell by nearly 80% over the same period, while the three largest cloud providers held 74% of the global cloud market and NVIDIA about 90% of the graphics processing unit market, and the five largest US technology firms are forecast to spend USD 660 billion on capital expenditure in 2026, up from USD 400 billion in 2025.

AI investment boom and macroeconomic effects

AI investment is already lifting capital spending, high technology trade and asset prices, while the supply-side benefits that would justify it remain uncertain.

The most pronounced macroeconomic effect of AI so far stems from investment in AI infrastructure. Capital expenditure on data centres, semiconductors, power facilities and related equipment is contributing measurably to GDP growth in the most exposed economies and is increasingly financed through bonds, private credit and securitisation rather than internal cash flow. Because most of the relevant high technology equipment is imported, the build-out is also raising imports and import prices, widening the US current account deficit and generating spillovers to technology exporters. Research characterises the current phase as a build-out in which investment, trade and asset prices respond before output does. The boom adds to demand and price pressures in the near term, while the productivity gains that would expand supply remain uncertain and are expected to arrive later. The Bank for International Settlements notes that this simultaneous effect on demand and supply obscures cyclical signals and complicates central banks’ assessment of underlying economic conditions. Several studies also warn that competition among developers and uncertainty about the size of the technology shock can push investment beyond the level that eventual returns would justify, as in earlier technology booms.

  • The Bank for International Settlements estimates that AI-related spending has reached around 1% of GDP in the most exposed economies, with industry projections putting global investment at USD 3 trillion to USD 4 trillion by 2030, up from USD 500 billion, and financing increasingly shifting towards private credit and bond markets.
  • Federal Reserve Board research shows that roughly 90% of the high-technology equipment used in the AI build-out is imported, mainly from East Asia, and that an investment-specific technology shock raises investment by about 2% over the first year, widens the current account deficit by about 0.2 percentage points of GDP and lifts import prices by about 1 percentage point, with spillovers concentrated in Taiwan, Korea and Mexico.
  • The Executives’ Meeting of East Asia-Pacific Central Banks observes that AI is already driving data centre investment, technology production and exports in the region while near-term productivity effects remain limited and uneven, and expects investment demand and higher technology import costs to add to inflation initially before lower production costs produce disinflationary effects over time.
  • The European Central Bank finds that euro area firms in the most AI-intensive sectors rely more heavily on equity and have shifted their debt financing from bank loans to market instruments, which may make their investment less responsive to monetary policy through bank lending and more sensitive to long-term yields.

Key sources