Metals in the age of AI: Game-changer or hyperscale hype?
GLD•AI data centers are metal-intensive, but their direct demand for copper may be limited: data centers could account for less than 0.9% of global copper use in 2035. Power infrastructure could lift copper and aluminium demand three or four times above estimates based only on facilities, while grid constraints and policy may delay construction.
1. Metals in AI facilities
Data centers consume an estimated 60 to 75 metric tons of metals per megawatt, while standard facilities use 5 to 10 megawatts and AI hyperscalers exceed 100 megawatts. The World Economic Forum estimates data centers will account for 6.0% of global gallium use, 2.4% of germanium and 1.2% of tin in 2035; copper's share is expected to be below 0.9%.
2. Power infrastructure matters
Estimates focused on data center buildings exclude metals used to supply their electricity. Wood Mackenzie estimates copper and aluminium demand could be three or four times higher when power connectivity is included, and projects annual power-capacity additions for data centers rising from 15–20 gigawatts today to 30–33 gigawatts in the early 2030s. It expects copper and aluminium use within facilities to peak in the early 2030s before declining.
3. Build-out faces constraints
An Oxford Smith School and Marex study says announced AI capacity may include “bragawatts” and that deployment is likely to be delayed, uneven and constrained. It found the average US grid-connection wait was four years between 2018 and 2023, compared with 18 to 24 months to build a data center. The column also points to political opposition and says gallium may benefit from the development of metal-based soft robots.
