Artificial Intelligence Doesn't Live Only in the Cloud: The Global Trade That Makes It Possible

Every time we get a response from an artificial intelligence (AI) system, the scene seems to take place in an immaterial space we call "the cloud." However, the cloud exists physically, occupies a place in space and time, and consumes materials and energy. Behind every interaction with AI there are microprocessors, servers, fiber optic cables, transformers, cooling systems, and a growing network of factories, suppliers, and infrastructure projects spread across the world. That physical world associated with AI creates a network of buyers and sellers that is changing the patterns of international trade.

In 2025, global exports of a selected set of 104 categories of AI-enabling goods reached an estimated value of $4.05 trillion, 25.6% more than in 2024. These products accounted for 16.6% of world merchandise exports and explained close to 40% of total nominal growth in global exports in 2025, according to estimates by the Inter-American Development Bank (IDB) that are consistent with the World Trade Organization (WTO) report for the first half of 2025. While the debate over AI's impact on various fronts remains open, the investment needed to build its infrastructure is already visible in global production and trade.

As part of its work generating knowledge on productive development and trade integration, the IDB studies these trends and offers analysis to help identify opportunities for the insertion of the Latin America and the Caribbean (LAC) region.

Trade statistics cannot determine whether a specific product was acquired for a data center, an AI lab, or some other use. What they do allow us to observe is the extent to which countries are increasingly producing and trading goods that are technically relevant to developing and operating the infrastructure that sustains this technology. For this reason, it is more accurate to speak of AI-enabling goods, a term used by the WTO to refer to products that support the production, deployment, and operation of these systems, even when they may also have alternative uses.

When thinking about the hardware behind AI, it is common to picture a state-of-the-art microprocessor, but the reality is more complex. The 104 product categories under the international six-digit Harmonized System tariff classification linked to building and operating AI infrastructure can be grouped into four families:

The composition of AI-enabling goods exports shows that intermediate inputs accounted for 72.9% of the value exported in 2025, while final equipment represented 25.9%. Raw materials and specialized chemicals together account for less than 1%, although they are critical links in the value chain. Without ultrapure silicon or noble gases, there are no semiconductors.

The AI value chain includes, in addition to semiconductors, network equipment, fiber optics, electrical panels, transformers, converters, pumps, ventilation and cooling systems, sensors, and measuring instruments. In short, language models are just the tip of the iceberg of a much larger industrial system.

The recent growth in trade of AI-enabling goods has clear leading players. Processing units, the heart of AI servers, more than doubled their global exports in a single year, adding $142 billion in 2025. Following are computer equipment parts and accessories, which include graphics cards and memory modules, which grew by 71% in 2025; data transmission and switching equipment that connects data centers to one another, which grew by 48%; and processors, controllers, and memory integrated circuits.

The increase also extends to less visible products, such as cables with connectors, printed circuits, electrical converters, and control and distribution panels, all essential for powering and cooling this infrastructure.

For LAC countries seeking to join this production chain, a fundamental question is how difficult it is to manufacture these goods. One way to approximate that difficulty is to assess the amount of productive capabilities, technical knowledge, and specialized suppliers their production requires.

The Product Complexity Index (PCI) captures precisely that idea. The indicator assigns a higher level of complexity to goods that only a small number of countries manage to export competitively and that, in addition, are produced by economies with highly diversified export structures. In other words, a product is more complex when it requires a combination of sophisticated productive capabilities that few countries have managed to develop.

Analyzing the PCI of each of the 104 tariff lines for AI-enabling goods shows that 94% of the value exported in 2025 corresponds to products of high or very high complexity. Manufacturing memory integrated circuits, semiconductor production machinery, or precision measuring instruments requires capabilities that today are concentrated in a small number of countries.

However, the picture is more heterogeneous than that aggregate figure suggests. The remaining 6% of global trade in AI-enabling goods, equivalent to more than US$200 billion, corresponds to medium- or low-complexity products, such as insulated cables and conductors, connectors, and cooling equipment like industrial fans and small transformers, segments that grew at double-digit rates in 2025.

Although it represents a minority share of AI-related trade, it is a market of enormous scale, and its value exceeds all goods exports by Chile and Peru combined in 2025. For economies still far from the technological frontier, these goods can serve as a gateway into the AI value chain. They represent activities where existing industrial capabilities allow countries to compete, build knowledge, take advantage of economies of scale, and gradually move toward higher-complexity segments.

LAC's participation in this chain is still limited, although it already shows concrete signs of insertion. In fact, our region is today a net importer of this infrastructure: in 2025 it exported $165.8 billion in AI-enabling goods and imported $227.5 billion, meaning it bought 1.37 times what it sold.

Mexico accounts for most of this insertion: it exported $154.1 billion in AI-enabling goods in 2025 and represents close to 93% of the region's exports of this group of products. In that country, momentum is already visible in specific manufacturing hubs: economic press reports point to Ciudad Juarez and Guadalajara as relevant nodes, with a cluster in Chihuahua associated with manufacturers such as Foxconn, Inventec, Pegatron, Wistron, Quanta, and Wiwynn, and a platform in Jalisco built on capabilities accumulated by companies such as Intel, IBM, HP, and Flex.

Costa Rica offers another, more limited but equally relevant example. Its exports of AI-enabling goods exceeded $4.3 billion in 2025, based on the basket of 104 products. Other LAC countries participate incipiently through critical minerals, energy capacity, or specialized manufacturing niches.

An emerging opportunity exists for LAC, although capturing it will not be automatic. It will depend on the productive capabilities each country has and can develop, its proximity to the goods that AI's expansion demands, and the policies that accompany the process.

The next installments of this blog series will look at who is demanding these products, how investment in the United States is redrawing the trade map, the case of Mexico as early evidence of possible "AI-shoring," and what governments and the private sector can do to turn this new demand into greater local value added, stronger productive capabilities, and well-being. Stay tuned.

Copyright © 2026 Inter-American Development Bank ("IDB"). This work is subject to the Creative Commons Attribution 4.0 International Public License CC BY 4.0. The terms and conditions indicated in the URL link must be met and the respective recognition must be granted to the IDB. Any dispute arising under this license that cannot be settled amicably shall be resolved in accordance with the following procedure. Pursuant to a notice of mediation communicated by reasonable means by either the licensee or the licensor to the other, the dispute shall be submitted to non-binding mediation conducted in accordance with the WIPO Mediation Rules. Any dispute that cannot be settled amicably shall be submitted to arbitration pursuant to the United Nations Commission on International Trade Law (UNCITRAL) rules. The use of the IDB name for any purpose other than the respective recognition and the use of the IDB logo are not authorized by this license and require an additional license agreement. Note that the URL link includes terms and conditions that are an integral part of this license.

The opinions expressed in this work are those of the authors and do not necessarily reflect the views of the IDB, its Board of Directors, or the countries they represent.