AI in a fragmenting world: Why ‘connector’ economies can still gain

How much of artificial intelligence’s economic promise will survive in a more fragmented global economy? Translating advances at the frontier into broad productivity gains depends on how widely AI-related technologies and inputs diffuse and are adopted across countries. Broad adoption is central to the macroeconomic payoff from AI (Filippucci et al. 2024), while economies that maintain diversified links across geopolitical blocs may play an increasingly important ‘connector’ role (Aiyar and Ohnsorge 2024). Bringing these two strands together raises a natural question: how does fragmentation affect the gains from AI, and what can economies outside the technological frontier do to preserve them?

In Eichengreen et al. (2026), we study this interaction using a multi-country, multi-sector model of trade and production networks. Our central message is simple. AI may be created at the technological frontier, but its economic gains can spread much more widely. Countries that do not develop frontier models can still benefit by importing AI-related hardware, software and services, using AI-enabled intermediate inputs, and participating in the infrastructure and supply chains that support AI deployment.

AI is global even when innovation is not

We distinguish three layers of the AI ecosystem: frontier innovators; supply-chain and infrastructure providers; and adopters. The first layer is highly concentrated, especially in the US, China, and the EU. The other two are much broader. This matters because firms can become more productive without developing AI itself: they can use better semiconductors, software, cloud services, or AI-enabled machinery produced elsewhere. This emphasis on international transmission complements work on countries' differing exposure and preparedness for AI adoption (Cazzaniga et al. 2024).

Our benchmark simulation using the IMF’s ECLIPSSE model illustrates the size of this diffusion channel. When AI raises productivity at the frontier and also improves the efficiency with which AI-related intermediate inputs are used elsewhere, modelled global GDP is 1.64% above the baseline. The gains are 2.28% in the US, 1.87% in China, 1.93% in the EU and 0.94% in the rest of the world. These numbers are scenario-based model estimates, not forecasts. Their purpose is to show how gains propagate through production networks.

Geoeconomic fragmentation can weaken this process in two ways. The first is through barriers to trade and technology diffusion: tariffs, export controls, investment screening, restrictions on data flows, and incompatible digital regulations can make AI-related inputs more costly or less accessible. These channels are part of the broader costs of geoeconomic fragmentation documented by Aiyar et al. (2023), while recent supply-chain reallocation shows how firms and countries respond when direct linkages become more difficult (Alfaro and Chor 2023).

The second channel, which adds to these trade and diffusion effects, can be more damaging. Fragmentation can also slow innovation at the frontier itself by reducing collaboration, knowledge flows, market scale and access to complementary inputs. In our simulations, trade fragmentation alone – the first channel – reduces global GDP gains from 1.64 to 1.49 percent. The effect on the rest of the world is small: its gain slips from 0.94% to 0.93%. But when fragmentation also slows frontier AI progress, global gains fall further to 1.29% and rest-of-world gains to 0.85%.

Figure 1 Fragmentation reduces the gains from AI, especially when it slows frontier innovation

The distinction between the two channels discussed above has an important policy implication. The main global risk is not simply that AI-related trade is rerouted. It is that fragmentation damages the innovation process that creates technologies for everyone.

Connector economies have options

For economies outside the AI frontier, fragmentation does not imply passivity. The MENAP economies in our sample provide a useful case. Many maintain significant trade and investment links with the US, Europe and China. That diversification can become an asset if geopolitical rivalry intensifies. At the same time, the region is far from homogeneous: AI preparedness varies widely, and average AI intensity in intermediate inputs remains below that of advanced economies and other emerging markets.

Our simulations compare ‘choosing sides’ with three more active strategies: acting as a connector, improving AI preparedness, and strengthening local AI-related capacity. A forced alignment with one bloc can be costly when it cuts countries off from important existing suppliers. By contrast, a connector strategy – lowering lawful trade and regulatory frictions with partners on multiple sides – can more than offset the losses from fragmentation in our experiment. Improving digital infrastructure, skills, and institutions also raises the capacity to absorb AI, while local software, data, adaptation, and deployment services can deepen domestic gains.

Figure 2 For MENAP, connector and local-capacity strategies can more than offset fragmentation losses

These exercises are conditional benchmarks, not cost-benefit estimates: they abstract from fiscal costs and implementation risks, and they do not assume circumvention of export controls. But the ranking of strategies is informative. Preserving diversified technology links generally performs better than rigid bloc alignment.

The broader lesson extends beyond the region. For countries that are unlikely to dominate frontier AI research, ‘optionality’ is an economic asset. Policies that keep trade and technology relationships open, improve interoperability, invest in human capital and digital infrastructure, and build local capacity to adapt and deploy AI can preserve access to innovation originating elsewhere.

AI has many characteristics of a general-purpose technology (Calvino et al. 2025), but its benefits will not diffuse automatically. In a fragmenting world, the architecture of international economic relationships may matter almost as much as the speed of innovation itself.

Author note: The views expressed are those of the authors and do not necessarily represent the views of the IMF, its Executive Board, or IMF management.

Aiyar, S and F Ohnsorge (2024), “‘Connector’ countries in a geoeconomically fragmented world”, VoxEU.org, 24 August.

Aiyar, S, J Chen, C H Ebeke, R Garcia-Saltos, T Gudmundsson, A Ilyina, A Kangur, T Kunaratskul, S L Rodriguez, M Ruta, T Schulze, G Soderberg, and J P Trevino (2023), “Geoeconomic Fragmentation and the Future of Multilateralism,” IMF Staff Discussion Note 2023/001.

Alfaro, L and D Chor (2023), “Global Supply Chains: The Looming ‘Great Reallocation’,” NBER Working Paper 31661.

Calvino, F, D Haerle and S Liu (2025), “Is generative AI a General Purpose Technology? Implications for productivity and policy,” OECD Artificial Intelligence Papers No. 40.

Cazzaniga, M, F Jaumotte, L Li, G Melina, A J Panton, C Pizzinelli, E J Rockall, and M M Tavares (2024), “Gen-AI: Artificial Intelligence and the Future of Work,” IMF Staff Discussion Note 2024/001.

Filippucci, F, P Gal and M Schief (2024), “Miracle or myth: Assessing the macroeconomic productivity gains from artificial intelligence”, VoxEU.org, 8 December.

Eichengreen, B, G Cui, A A ElGanainy, Y Korniyenko, E Shojaei, L Zeng, and F Zhang (2026), “AI in a Fragmenting World,” NBER Working Paper 35597.