While AI hardware and software assembly draws on labour and resources distributed across many countries, the financial returns flow predominantly to a handful of companies that hold the underlying patents and direct the technology's architectural development. Financial gains tied to the artificial intelligence expansion are concentrated at the upper tiers of the supply chain, UNCTAD has found.
According to WTO projections, AI-related chips and server hardware accounted for roughly one in six goods crossing international borders in 2025, driving nearly two-fifths of the sector's expansion when measured by value. Building an AI server draws on a wide-ranging network of industries — from the extraction of raw materials and the fabrication of semiconductors through to the manufacture of power infrastructure and thermal management equipment.
The benefits a country draws from a production chain depend heavily on which segments of that chain it can actually participate in. Raw materials for AI hardware flow from Chile and Peru, while the manufacturing chain advances through South Korea and Taiwan for semiconductor production, before reaching Malaysia and Mexico, where finished devices take their final form.
Raw materials such as silicon, copper, and tin — the physical inputs behind server hardware built to run AI workloads — made up just 0.2% of what those systems ultimately sold for, UNCTAD data showed. A breakdown of the Nvidia GB200 NVL72 server system's manufacturing chain, conducted by UNCTAD, brought the underlying disparity into view.
Estimates place the worth of the system at approximately $3.3 million. Of the roughly $2.92 million brought under the study's scope, the breakdown traced which activity types and supplier clusters drove more than 87 cents of every dollar in generated value — with 68 cents of that dollar remaining as net return once operating expenses and tax obligations were stripped away.
According to the findings, nearly all of the profit margin ended up concentrated at Nvidia, a company whose business model centres on chip architecture and design while leaving the actual fabrication to outside manufacturers. Of the total value that could be traced through the supply chain, workers directly engaged in physical production of the server system received just 4.3 cents of every dollar.
Overhead roles — corporate managers and design engineers among them — claimed a 9.9% slice, pushing the total labour portion attributable to workers across the finished product to 14.2%. According to the report, nations supplying raw materials saw their mineral exports climb on the back of artificial intelligence expansion, yet their portion of the value generated along the supply chain has not grown accordingly.
Assembly and manufacturing work in lower-income economies does generate employment, yet the financial gains tied to those activities tend to fall short of what flows from technology design at the higher end of the value chain. A report released by UNCTAD on Oct. 9 — titled "Trade and Development Report 2026" — called on nations still building their economies to place greater weight on nurturing homegrown supply chains and developing a capable workforce, rather than concentrating efforts on drawing in capital from abroad.
Gaining a foothold in the upper tiers of AI component production remains out of reach for many lower-income countries, where the combination of substantial upfront capital requirements and geopolitical constraints on technology access continues to narrow the path forward.
