The AI Arms Race Drives Up Copper Demand, but Uncertainty Looms Over the Market

加工费回升难改困境,中国铜冶炼厂利润短期改善
Published on: Jun 5, 2026
Author: Amy Liu

According to data, a cryptocurrency data center with a megawatt of installed capacity requires 21 metric tons of copper, while an AI training data center in China has a copper intensity as high as 47 tons per megawatt. The escalating global AI arms race will undoubtedly drive up demand for copper. But by how much exactly?

S&P Forecast: AI Will Drive a 50% Increase in Copper Demand by 2040

S&P Global forecasts that copper usage in data centers and related infrastructure will rise from 1.1 million tons in 2025 to 2.5 million tons in 2040.

However, beneath this forecast lie numerous “uncertainties.” Demand could be as high as 2.7 million tons or as low as 1.7 million tons, depending on the interplay of several rapidly changing variables. As the author of the report “Copper in the Age of AI” puts it: “This wide range highlights both the uncertainty and the scale of the challenges ahead.”

More Power, Less Copper

Quantifying how much copper a data center uses is also a rapidly moving target. The battle for AI supremacy is essentially a battle for computing power, and as chips evolve, the architecture of rack designs and cabling must change accordingly.

S&P Global notes that interconnect cabling between processor racks has already begun shifting from copper to fiber optics. This could reduce copper intensity by 4 to 5 tons per megawatt of installed capacity—a significant change, given that non-cryptocurrency data centers use between 30 and 40 tons of copper per megawatt.

More importantly, chip company Nvidia believes that even copper struggles to meet the low latency and high bandwidth required by next-generation AI centers. Using conventional low voltage would require an “unsustainable number of copper cables.” Nvidia has proposed transitioning to 800 volts, allowing the same wire gauge to carry 157% more power. Simpler setups also mean fewer copper conductors and smaller connectors. Reducing copper usage is both a cost consideration and a key pathway to achieving higher rack power density.

Pricing Complexity

All emerging industries are prone to a disconnect between promise and delivery, but power availability, grid connections, and critical metal supplies are structural, real-world constraints on the AI revolution. While no one will stop using copper in data center designs, the intensity of its use depends on the race for higher computing power, which in turn demands continuous evolution in AI architecture.

Copper as a new demand vector in data centers contains multiple dynamic variables. And the market, which sees AI as the next major boon, may not have fully priced in this complexity.

Summary: Copper plays a key role in the growing demand from AI data centers, but actual usage is constrained by multiple factors, including grid bottlenecks, technological substitution (such as fiber optics and high-voltage solutions), and supply tightness for other metals like germanium and gallium. The certainty of future copper demand is far lower than current market enthusiasm suggests.

Base Metals Copper Energy Metals Mining