## AI Can Lift Global Growth

## Source details

**Canonical URL:** [AI Can Lift Global Growth](https://www.imf.org/-/media/files/publications/fandd/article/2026/03/estevo.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/fandd/article/2026/03/estevo.pdf.md)
- [Structured JSON version](/-/media/files/publications/fandd/article/2026/03/estevo.pdf.json)

---

### AI-driven investment and growth patterns
- AI-related investment "now accounts for a large share of GDP growth" in the United States and is fueling demand for servers, data centers, software, and power infrastructure.
- The US economy has entered a "two-speed expansion": AI-intensive sectors are racing ahead while construction, manufacturing, and interest-rate–sensitive industries fall behind.
- Bureau of Economic Analysis: "investment in information-processing equipment and software grew by 16.5 percent from a year earlier in the third quarter of 2025."
- Global growth remains uneven: Europe and Japan have stabilized but remain dependent on loose monetary policy; emerging markets benefit from lower yields and a weaker dollar but are increasingly reliant on technology-related investment and capital inflows.
- Estimation: "Data centers worldwide may require $6.7 trillion in capital expenditure by 2030."

### Measurement challenges: "Measuring immeasurables"
- National accounts were designed for an industrial age and poorly capture intangible capital now central to value creation (data, algorithms, proprietary models, cloud infrastructure).
- Official statistics capture some of the shift (software and R&D) but miss many drivers of productivity:
  - Costs of training large models, refining datasets, and creating new applications are often expensed rather than capitalized.
  - Semiconductors are frequently treated as intermediate goods rather than carriers of embedded intellectual property.
- Statistical paradox: GDP can simultaneously overstate AI’s immediate contribution (by counting massive capital outlays) and understate broader productivity spillovers.
- Policy risk from measurement lag: misreading apparent slack or inflation pressures could lead to inappropriate tightening or easing. Example: "The Federal Reserve... faces a more complex policy landscape."

### New geographies, trade, and capital flows
- AI is reshaping trade and capital-flow patterns: imports and exports of computers, servers, and semiconductors have surged; manufacturing and assembly are shifting toward Southeast Asia, India, and specialized US hubs such as Texas and the Gulf Coast.
- Re-regionalization rather than de-globalization: US and China remain dominant; Europe pursues industrial policy and investment incentives to catch up.
- Digital participation vs digital depth (Institute of International Finance research):
  - Digital participation = use of imported digital tools.
  - Digital depth = ability to produce and export digital goods/services and embed them in domestic value chains.
- Emerging markets with digital depth (China, India, Korea, and specialized hubs) attract more stable FDI tied to AI-era production; others remain consumers of imported technologies and rely more on volatile portfolio flows.
- Equity markets favor hyperscalers: a small group of tech giants accounts for a disproportionate share of global AI-related capital expenditure and productivity expectations.

### Energy, grids, and supply-side constraints
- The scale of computing power for AI training and inference makes electricity generation and grid capacity "critical macroeconomic variables."
- Energy bottlenecks could:
  - Delay AI diffusion.
  - Anchor a higher level of core inflation.
  - Generate localized overheating while other sectors remain weak.
- Grid investment is becoming a central supply-side constraint, blurring industrial and macroeconomic policy responsibilities.

### Diffusion versus concentration: scenarios and implications
- Historical lesson: general-purpose technologies raise aggregate productivity only after years of complementary investment in skills, management, and institutions (electricity and IT took decades).
- Two broad scenarios:
  - Concentration scenario: AI adoption remains concentrated among hyperscalers and specialized providers → returns plateau after the investment cycle peaks; economy remains capital-heavy and employment-light.
  - Diffusion scenario: AI spreads across industries with complementary investments → potential for sustained lift in potential output.
- Current evidence: corporate surveys indicate diffusion is underway but uneven; many firms experiment with AI while a smaller group implements it at scale.
- Risks: inadequate infrastructure and outdated statistics can make diffusion collide with supply constraints and policy lag, producing volatility as growth oscillates between investment bursts and adjustment pauses.

### Policy implications and recommendations
- Measurement:
  - Evolve measurement frameworks to capture intangible capital (data, models, cloud infrastructure).
  - Reconsider capitalization practices for costs tied to model training, dataset refinement, and application development.
- Fiscal and monetary policy:
  - Account for sectoral divergence and new supply constraints (energy and grid limits) when setting policy.
  - Avoid misinterpreting capital-heavy investment outlays as purely demand-driven inflation.
- Structural and industrial policy:
  - Support complementary investments in skills, management practices, and institutional adaptation to promote diffusion.
  - Europe should retool industrial and digital policy to avoid missing out.
  - Emerging markets should balance using cheap energy or favorable regulation with long-term competitiveness.
- International cooperation:
  - Coordinate to ensure AI diffusion benefits are not confined to a few economies.
- Monitoring:
  - Treat digital depth as an underappreciated channel affecting capital flow dynamics, akin to fiscal credibility or exchange-rate regimes, and monitor it closely.

*marcello estevão is managing director and chief economist at the Institute of International Finance and a professor at Georgetown University.*

---


_Source: https://www.imf.org/-/media/files/publications/fandd/article/2026/03/estevo.pdf_