---
id: "lib-2026-05-00-ai-and-trade-europe-adoption"
type: "library"
title: "AI and trade: Europe adoption"
lang: "EN"
org: "Harvard Growth Lab"
source_type: "Paper"
date: "2026-05-01"
week: "2026-W18"
url: "https://nosov.ai/library/lib-2026-05-00-ai-and-trade-europe-adoption/"
original_url: "https://growthlab.hks.harvard.edu/wp-content/uploads/2026/05/econpol-forum-2-2026-yildirim-AI-and-trade.pdf"
tier: "A"
tier_name: "Strong evidence"
tier_reason: "Harvard Growth Lab economist Muhammed Yıldırım summarises general-equilibrium trade-model results from an OECD AI paper co-authored with OECD productivity economists, a peer-level method applied to a Europe-focused policy question."
summary_basis: "full-text"
original_minutes: "15"
tags: ["ai adoption", "trade", "europe", "productivity"]
frames: ["solow-paradox"]
---

# AI and trade: Europe adoption

Tier A (Strong evidence): Harvard Growth Lab economist Muhammed Yıldırım summarises general-equilibrium trade-model results from an OECD AI paper co-authored with OECD productivity economists, a peer-level method applied to a Europe-focused policy question.

Summary based on the full text.

## In short

A short policy article in EconPol Forum 2/2026 (ifo Institute) by Muhammed Yıldırım, director of academic research at the Harvard Growth Lab and associate professor at Koç University. It extends the OECD paper "AI Meets Trade" (Filippucci, Gal, Laengle, Schief and Yıldırım, 2026) with Europe-specific simulations. The finding: countries gain from foreign AI progress through cheaper imports, but those gains do not offset lost competitiveness if domestic adoption lags. If the EU falls behind, its share of global income drops by an estimated 2.5% to 4.1% in ten years.

## Key numbers

- 2.5 percent to 4.1 percent: estimated fall in the EU's share of global income after ten years if it lags trading partners on AI adoption
- 14 percent and 55 percent: range of experimental productivity gains on AI-exposed tasks used as model inputs
- close to one percentage point: upper end of projected annual productivity growth gains, depending on industry mix and adoption pace

## What it reports

- Micro inputs: productivity gains on AI-exposed tasks of 14% to 55% from experimental studies (Brynjolfsson, Li and Raymond 2025; Gambacorta et al. 2024).
- Largest sector gains projected for information and communication, finance and insurance, and professional and scientific services; smallest for agriculture, construction, and accommodation and food.
- Projected increases in annual productivity growth run from modest to close to one percentage point, depending on industry mix and adoption speed.
- Luxembourg, Sweden, Belgium and Ireland are positioned to gain most; Cyprus, Bulgaria, Romania and Croatia fall below the world average even with medium adoption.
- In welfare terms Ireland leads the EU and exceeds the US; its gains from foreign adoption are almost as large as its domestic gains. France's foreign component is small.
- Two trade channels: a passive import-price channel that benefits everyone, and a conditional competitiveness channel that rewards only countries adopting at or above their competitors' pace.
- With symmetric adoption the EU's global income share stays broadly stable; with EU lag it falls by 2.5% to 4.1%.
- Three policy priorities: EU-level coordination on infrastructure and skills, open trade and digital integration, and sector-aware diffusion policy with better firm-level adoption data.

## How they got it

Task-level productivity estimates are combined with sector exposure and country adoption rates to produce country-sector productivity shocks. These feed a general-equilibrium world trade model (Baqaee and Farhi 2024). Adoption follows S-curves calibrated to electricity (slow), PCs and the internet (medium) and mobile phones (fast); results shown use slow and medium. Welfare gains are decomposed into domestic, EU, non-EU and competition effects over ten years.

## Limits

- Results are model simulations; they depend on the experimental gain estimates and on assumed diffusion curves.
- The ten-year horizon and adoption scenarios are stylised; the article itself notes that firm-level adoption data in Europe is limited.
- It weighs the economic cost of slower adoption but does not quantify benefits of the EU AI Act.

## Why it matters for companies

For a European firm in a knowledge-intensive or export sector, the model says buying AI-enabled inputs from abroad brings cheaper inputs but not protection: competitors that adopt faster take market share. The sectors with the highest gains (finance, professional services, ICT) are also the ones most exposed to that pressure. Adoption pace inside the firm, not access to frontier models, is the variable that decides relative position. Companies in Southern and Eastern Europe face a wider gap by default and cannot count on trade spillovers to close it.

Original: https://growthlab.hks.harvard.edu/wp-content/uploads/2026/05/econpol-forum-2-2026-yildirim-AI-and-trade.pdf