AI-driven copper boom 'overhyped'

The AI boom may only give copper demand a slight lift if new datacenters prove less metal-intensive than expected, analysts told S&P Global Commodity Insights.

Datacenter and AI copper bulls have projected that demand for the metal will skyrocket as power-hungry and metal-intensive computers proliferate. Commodities trader Trafigura Group Pte. Ltd. said in April that annual copper demand could rise by as much as 1 million metric tons by the end of the decade. Likewise, J.P. Morgan analysts have projected that new demand could require as much as 900,000 metric tons of copper per year, while Bloomberg Intelligence sees global copper demand jumping by 2 MMt per year by 2030.

But some analysts caution that innovation could drive down the amount of copper that datacenters will actually need at each site.

"There has been a lot of hype about additional copper demand driven by AI datacenters in recent months," Ruilin Wang, an analyst at S&P Global Commodity Insights, said in an email, while noting that "some estimates exaggerate."

AI potential

Analysts agree that new AI tools such as ChatGPT are propelling datacenter growth and power consumption. The International Energy Agency expects global electricity demand to double by 2026 to more than 1,000 TWh, and Goldman Sachs sees datacenter power demand growing 160% by the end of the decade compared to 2023.

Some AI tools are also processing and power hogs. As ChatGPT chews on user queries, it consumes nearly 10 times more electricity than simple Google searches, according to the Electric Power Research Institute, a US utility-funded research center, as well as others.

Larger, more power-hungry datacenters require more copper in the computers themselves and also in the web of wires and cables that support them.

Bloomberg Intelligence and J.P. Morgan use similar copper utilization assumptions, projecting that every megawatt of datacenter power capacity will require 20-40 metric tons of copper. Thus, the analysts see one boom — datacenter and power growth — leading to another: copper demand. If datacenter power consumption grows at a 20% compound annual growth rate, annual copper demand could increase by 400,000-900,000 metric tons by 2030, J.P. Morgan analysts said in a March 28 report. In a base case using 15% growth, annual copper demand could jump by 250,00-500,000 metric tons by the decade's end, the analysts said.

Others, including Trafigura and Bloomberg, see copper demand going even higher.

For context, global mined copper production totaled an estimated 22.6 MMt in 2023, according to S&P Global Market Intelligence data. An additional 2 MMt would be the equivalent of adding about two more mines like Chile's Escondida, the largest copper operation in the world. The fresh demand is also projected to come amid copper supply deficits set to deepen in the late 2020s and into the 2030s.

Too intense!

However, some experts doubt AI will fuel as big a run-up in copper demand as projected by Trafigura, Bloomberg, J.P. Morgan and others.

"I think the 1-2 million metric ton [estimate] for AI-related [demand] is on the high side," Colin Hamilton, BMO Capital Markets' managing director for commodities research, told Commodity Insights.

Goldman Sachs, while bullish on copper overall, also doubts that AI and datacenters will drive a surge in copper demand.

"We do not think it is significant for the time being," Adam Gillard, part of Goldman Sach's commodities institutional sales team, said during a May 22 podcast. "Over the next three years, we think there could be an additional 100,000 [metric] tons of copper demand from datacenters, AI. So that's around 3,000 [metric] tons per month. And to put that into context, the Chinese consume about 1.2 million [metric] tons per month."

Two key assumptions — growth in datacenter power consumption and datacenter copper intensity — account for much of the wide range in the forecasts.

Macquarie analysts estimated that the AI boom will use far less copper than some researchers expect, partly because of an ongoing shift in computer-related infrastructure away from copper-rich cable to fiber optic cable. Taking this trend into account, they projected the copper intensity of a modern datacenter at 4.2 metric tons per megawatt, which is 10.5% to 21% of the estimates by J.P. Morgan and Bloomberg Intelligence.

"The amount of copper cables used in datacenters for data transfer has decreased as computing power has gone beyond copper's technical limit of 40 [gigabytes per second] over short distances," the Macquarie analysts said in a May 14 report, referring to data transfer speeds. Meanwhile, "data speeds are increasingly moving towards transmission rates of up to 500 [gigabytes per second] with fiber now the preferred choice to connect servers, switches and racks with ranges of a few to hundreds of meters."

Macquarie analysts expect datacenter and AI-related copper demand to grow by just 200,000 metric tons per year by decade's end. "We think the potential impact of [datacenters] and AI on copper demand, with reported estimates ranging from [500,000 metric tons per year] to over [2 MMt per year] by 2030, has been overhyped," the analysts said in a May 23 report.

AI evolution

How the world uses new AI tools and what gains in power efficiency come through innovation will in part dictate the material intensity of datacenter growth. Already, lead AI chipmaker NVIDIA Corp. announced a new chip in March that it said would be up to 25 times more energy efficient than current technology.

New cooling systems are one area the industry is assessing for innovation to drive down energy use and costs, Sajjad Moazeni, an assistant professor in the University of Washington's electrical and computer engineering department, told Commodity Insights. And as AI development matures, chatbots that sacrifice some accuracy for much lighter power consumption could become more popular, Moazeni said. AI could also propel energy efficiency in other sectors, including transportation and power distribution, offsetting overall energy and material demands.

"It's very, very hard at this point to figure out what exactly are going to be the overall, macro-level changes," Moazeni said.

Commodity Insights' Wang also said fresh copper demand from AI is coming off a low base and that consumption could moderate after an initial surge. Taking note from other industries, Wang said copper intensity in sectors such as electric vehicle manufacturing has tended to decline as the technology advances.

"So a research result ... this year may not be valid three to four years later," Wang said.