The Global Trade Effects of the AI Infrastructure Boom

February 13, 2026

The Global Trade Effects of the AI Infrastructure Boom

François de Soyres, Alex Haag, Mike Liu and Eva Van Leemput¹

Artificial intelligence (AI) has become a key driver of the global economic outlook, underscored by the unprecedented scale of announced investment commitments aimed at expanding AI-related infrastructure. The AI boom is also increasingly influencing international trade by boosting demand for critical inputs and intermediate goods needed to build data centers. This surge in AI-related investment has already supported international trade since early 2025, with strong demand for critical components boosting exports in key supplier economies.

The Surge in AI Infrastructure

AI-related capital expenditure has accelerated sharply in recent years with newly announced projects pointing to a sustained and substantial buildout ahead. For example, U.S. data-center spending alone is expected to exceed half a trillion dollars in 2025. As documented in Haag (2025), the U.S. leads globally in terms of the AI infrastructure build-out and planned investments, followed by China, with other economies, notably Europe, lagging. The U.S. and China drove a massive expansion in AI-related computing capacity in 2024 (proxied by GPU cluster performance), with the U.S. surging higher through the first half of 2025 (Figure 1A).

AI-related computing capacity is expected to increase dramatically through 2030, as illustrated by planned expansions in Figure 1B.² According to one estimate from Epoch AI, these projections suggest that the boom in AI-related computing capacity is set to accelerate further in the coming years. According to these data, the U.S. is expected to maintain a leading position, but several other economies-particularly in Asia and the Middle East-are also planning notable expansions.

Rising data center construction has boosted global demand for AI-related equipment and high-tech inputs. According to the WTO's October 2025 Global Trade Outlook (PDF), AI-related trade drove nearly half of merchandise trade growth in the first half of 2025, despite representing only about 15 percent of total trade. Much of this surge is tied to equipment and tools for semiconductor manufacturing, meaning that broadly defined AI-related trade may somewhat overstate the share of trade directly supporting frontier AI applications. In the next section, we provide a narrower definition of AI-related trade to better isolate the components that truly reflect the AI boom.

Defining AI-Related Trade

Definitions of AI-related international trade vary widely. Under the WTO classification, AI-related products encompass roughly 100 product lines, ranging from raw silicon and semiconductor manufacturing equipment to high-end servers that power AI applications. While these categories are useful for identifying AI-related goods, they also include a substantial share of non-AI products associated with broader digitization trends. Accurately measuring AI-related trade is challenging due to several factors:

• HS-6 limitations: Most international trade data are only available at the HS-6 level, which both obscures smaller AI-related subcategories and mixes AI with non-AI products. For example, including codes for personal laptops (HS 8471.30) may overstate AI-related trade, while the code for integrated circuits (HS 8542.31) contains a mix of AI and non-AI components
• HS code reporting nuances: Some HS codes may not fully capture actual trade flows. For example, HS 8542310040, the code for Graphics Processing Units (GPUs), shows U.S. import volumes far below what would be expected based on NVIDIA chip production by Taiwan Semiconductor Manufacturing Company (TSMC). One plausible explanation is that GPUs are often embedded in more complex products-such as graphics cards or server assemblies-before being exported, rather than being shipped as standalone components.
• Inconsistent national reporting. Countries may classify the same product differently. For example, the US and Taiwan reported roughly $13 billion in trade in July 2025 of HS 8471.50, 8471.80, and 8473.30 combined (known AI-related products). However, the US reported that the majority of this was HS 8471.50, while Taiwan reported that 8471.80 and 8473.30 were higher than 8471.50 (see Appendix).

To address these limitations, we develop a narrower, purpose-built classification of AI-related trade that focuses on components directly tied to the buildout of AI computing capacity. Specifically, we concentrate on three HS-6 categories covering servers, graphics cards, and related parts essential for AI data center tasks, including training and inference (Table 1). Aggregating these codes helps exclude high-tech goods less relevant to frontier AI development and partially addresses the classification challenges noted above.

Table 1: AI-Related HS Codes

While this classification aims to capture the core hardware components associated with the AI infrastructure buildout, it is not without some limitations. On the one hand, it may underestimate AI-related trade by omitting relevant products classified under other HS categories. On the other hand, because the selected HS codes can also include non-AI products, the resulting measures may overstate the trade flows that are strictly attributable to AI-specific activity. In the following sections, AI-related trade refers specifically to trade in these three HS-6 codes.

Global Trade in AI-Related Goods

The recent surge in AI-related investment has fueled strong growth in demand for our definition of AI-related products, with more than 272 billion dollars in trade for these products in the first half of 2025, a 65% increase over the first half of 2024. Figure 2A shows that since 2024, AI-related imports have more than doubled, with sustained growth in 2024 and a massive surge in imports in 2025: Q1, partially due to frontloading. Since then, imports have largely traveled sideways through July 2025 at a sustained high level.

Looking at individual countries, the U.S. and China drove growth in AI-related imports in 2024, with the U.S. accelerating further in 2025, while Chinese imports, after an initial surge in early 2025 likely attributable to frontloading, fell in the second quarter of 2025 amid geopolitical tensions and restrictions on global exports of high-tech goods (Figure 2B). Mexico, which largely assembles high-tech hardware for the U.S. market, has experienced significant growth in AI-related imports from a low level, a significant portion of which come from Taiwan. By contrast, AI-related imports from the European Union and other foreign economies have grown only slightly over the past few years. Taken together, the data since 2025: Q1 point to AI-related imports to the U.S. offsetting declines in China and Hong Kong, while imports in other countries are growing only slowly and from low levels.

Figure 3 shows that trade in AI-related goods has expanded more rapidly than global merchandise trade since 2018. While total world imports have increased at a more moderate pace, AI-related imports accelerated markedly from 2023 onward, reflecting strong demand associated with the construction of data centers and the buildout of AI infrastructure. Although AI-related goods still account for a relatively small share of global trade, their importance has risen noticeably, from about 1 percent of total merchandise trade in 2018 to roughly 2 percent in 2025. This surge has supported upstream suppliers and economies specializing in high-tech hardware assembly. At the same time, the resilience of global trade observed since early 2025 also reflects other forces, notably China's continued expansion in non-U.S. markets, which is largely unrelated to AI. Together, these dynamics illustrate how new technology-driven sectors can provide a countervailing boost to merchandise trade amid broader headwinds, as also discussed in Kose et al. (2025).

AI Demand Boosting Growth Abroad

Robust demand from the United States-and, to a lesser extent, China-for AI-related products has driven a sharp increase in exports from several export-oriented economies, notably Taiwan, Mexico, and Vietnam. The left panel of Figure 4 shows that exports of AI-related products to the United States surged in 2024 and 2025, with Taiwan playing a particularly prominent role: in the second quarter of 2025, Taiwan's AI-related exports to the United States reached roughly 14 percent of its GDP, reflecting its central position in the production of high-end AI chips through firms such as TSMC).

The right panel of Figure 4 shows exports of AI-related products to the rest of the world. For Taiwan, these exports also increased through 2025:Q1 but declined in 2025:Q2, at a time when shipments to the United States continued to expand. This pattern may partly reflect a reorientation of exports toward the U.S. market, with some U.S.-bound shipments potentially coming at the expense of exports to other destinations, although other factors may also be at play. By contrast, Mexico's AI-related exports are overwhelmingly directed toward the United States, while Vietnam's export growth is more broad-based, with sizable increases to both the U.S. and other foreign markets in the first half of 2025.

Looking ahead, AI-related trade is likely to remain an important feature of the global trade landscape as investment in AI infrastructure continues to scale. The buildout of data centers and the associated demand for specialized hardware are set to shape trade flows and support growth in key supplier economies, even as global trade faces broader headwinds. As a result, continued attention to AI-related trade will be essential for understanding how technological change is reshaping international trade patterns and influencing global growth dynamics.

References

Haag, Alex (2025). "The State of AI Competition in Advanced Economies," FEDS Notes. Washington: Board of Governors of the Federal Reserve System, October 6, 2025.

Kose, Ayhan, Alen Mulabdic and Collette Wheeler, 2025, "Global trade has remained resilient so far, but a sharp slowdown is underway", World Bank Blog, July 7, 2025.

Pilz, Konstantin, Robi Rahman, James Sanders, Lennart Heim, 'Data on GPU Clusters'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/data/gpu-clusters' [online resource]. Accessed 22 Dec, 2025.

United Nations Department of Economic and Social Affairs / Statistics Division. United Nations Commodity Trade Statistics Database, http://comtrade.un.org/db/.

World Trade Organization. "Global Trade Outlook and Statistics: October 2025 PDF." World Trade Organization, October 7 2025.

Appendix

1. François de Soyres, Alex Haag, Mike Liu and Eva Van Leemput are with the Board of Governors of the Federal Reserve System. The views expressed in this note are our own, and do not represent the views of the Board of Governors of the Federal Reserve, nor any other person associated with the Federal Reserve System. Return to text

2. Data on GPU Cluster Performance represent the sum of 16-bit O/Ps of all systems found within Epoch AI's database. This cluster performance from Epoch AI represent estimates of computing capacity for planned systems and should not be viewed as a forecast of AI-related computing capacity, either by the authors or the Board of Governors of the Federal Reserve System. Data are subject to updates on the details of planned systems as they become available. Return to text