Artificial intelligence supply chains are highly concentrated among a small number of companies and countries, leaving entire regions and sectors dangerously exposed to geopolitical friction and shortages of critical goods, trade credit insurer Atradius has warned.
Trade in AI-enabling goods drove around a third of global trade growth in 2025, according to new Atradius analysis of World Trade Organization data. But production of key AI hardware and the supply of critical mineral inputs are dominated by a small number of firms and economies, the insurer said, leaving global supply chains exposed to disruption from tariffs, export controls and logistical bottlenecks.
China is the largest exporter of AI-enabling goods, followed by Hong Kong, Taiwan and the US, Atradius found. But Taiwan and Hong Kong have the highest relative exposure, as AI-enabling goods account for 65% of Taiwan’s total exports and 56% of Hong Kong’s, higher than in China (18%) or the US (13%).
Atradius’ research found that “as geopolitical tensions rise and governments increasingly use technology policy as a strategic tool, disruptions to AI-related trade flows may become more likely”.
“In this respect, Europe appears relatively weakly positioned between Asia that dominates chip production and exports of rare earths, and the US that controls the software, cloud infrastructure, and chip design,” its Concentration risks in the global AI supply chain report noted.
Kyle Kong, Atradius underwriter for Asia and Oceania and the insurer’s trade sector specialist for electronics and ICT, told GTR that AI servers and data centres would be “the first sector to feel the pain” in a disruption scenario.
“AI servers and data centres are built around a small number of critical components, such as advanced GPUs, AI chips and high-bandwidth memory,” he said.
“The supply of these components is concentrated on a handful of manufacturers” and “generally, the inventory of the components is limited”, Kong added.
The automotive and industrial electronics sectors are also among the “most exposed in the near term” because the chips they rely on “are not held in large inventories”, Theo Smid, senior economist at Atradius, said.
Smid pointed to the Nexperia case in late 2025, when a Dutch-Chinese ownership dispute halted chip exports from China and pushed European carmakers to within weeks of production stoppages. The episode showed “how fast disruption in a relatively straightforward product segment can cascade through just-in-time manufacturing”.
The bigger long-term concern is the supply chain behind advanced AI hardware, Smid said, which includes “chokepoint” products such as lithography components, advanced logic chips as well as critical minerals like gallium
“In the upstream part of the supply chain, one of the tightest bottlenecks is in bringing different components together into a working AI chip.
“Advanced packaging is currently the tightest constraint: a single company (TSMC) controls roughly 90% of this capacity. High-bandwidth memory, dominated by three suppliers (SK Hynix, Samsung, Micron), is a related chokepoint.
“Disruption here would take longer to materialise but would be more severe and harder to substitute,” he added.
Smid said concentration risk would “generally feed into standard risk factors” for credit insurers, including counterparty and geographic concentration, political and regulatory risk, and event risk.
Kong said he would expect Atradius’s underwriting appetite for AI hardware exports “to differentiate more carefully within the sector rather than reduce on all fronts”.
“Likely we would be more favourable on leading manufacturers in the ecosystem. Potentially we would be more cautious to companies that have high concentrations on their suppliers, buyers and regions,” he said.
Efforts to increase diversification and protect global manufacturers are underway, with the EU’s Chips Act 2.0 aiming to cut single-source dependencies through a supply chain visibility platform, mandatory risk assessments in exposed sectors – including proposals for dual sourcing in automotive – and more state aid across the value chain.
The US has also launched Project Vault, a strategic minerals stockpile initiative, and signed critical mineral deals with Australia, Saudi Arabia and Ukraine, among others. Australia and Canada have also invested public funds in reserves and processing capacity.
Traders and financiers should “map exposure across tiers of the supply chain, not just direct counterparties, favour diversification in suppliers even if it comes at a cost premium, and factor in political and export control risk explicitly in financing and insurance decisions, given that the trigger for disruption is now as likely to be regulatory or geopolitical as commercial”, Smid said.
“Longer contract tenors and capital-intensive investments in AI-related capacity make this risk assessment even more important.”
