🐯虎嗅•Feb 26, 2026Stalecollected in 11m

AI Demand Diffuses in Fragile Supply

Post LinkedIn

🐯Read original on 虎嗅

#commodities #supply-chain #capexai-driven-commoditiesnvda copper

💡AI boom strains global supply chains—key signals for infra cost forecasts

⚡ 30-Second TL;DR

What Changed

AI Capex grows faster than overall Capex, concentrating demand

Why It Matters

Rising AI infra demand could inflate commodity costs, pressuring data center builds and AI expansion timelines.

What To Do Next

Track CFTC copper holdings data weekly to forecast AI infra cost pressures.

Who should care:Founders & Product Leaders

Key Points

•AI Capex grows faster than overall Capex, concentrating demand
•Copper rally needs supply/demand narrative and rising holdings at $13k price
•Fragile supply from power shortages in SE Asia/Africa, fastest builds in China
•Monitor Nasdaq, Korea inflows, CFTC holdings for trading signals

🧠 Deep Insight

Background and context from public sources — not the original article. 7 sources cited.

🔑 Enhanced Key Takeaways

•AI data center copper consumption is projected to reach 572,000 tonnes in 2028 (peaking mid-decade) before moderating, with cumulative locked-in copper surpassing 4.3 million tonnes by 2035—creating a structural supply constraint independent of broader electrification trends[2].
•Global copper production will peak in 2030 at 33 million metric tons, then decline, while demand reaches 42 million metric tons by 2040, creating a 10 million metric ton deficit (25% below demand) unless mining capacity expands significantly[1].
•Defense spending and humanoid robotics represent emerging demand vectors: defense demand alone is expected to triple by 2040, while 1 billion humanoid robots would require 1.6 million metric tons of copper annually (6% of current demand)[1].
•The International Copper Study Group projects a refined copper shortfall of 150,000 tonnes in 2026, reversing previous surplus forecasts, with JPMorgan expecting prices to reach $12,500/ton in Q2 2026 and UBS projecting $13,000 by year-end[3].
•China absorbs approximately 60% of global refined copper, with power infrastructure accounting for over 60% of demand growth through 2030, making Chinese manufacturing data and U.S. hyperscaler buildouts primary price direction indicators[4].

🛠️ Technical Deep Dive

•Hyperscale AI data center copper requirements: 40,000–50,000 tonnes per major facility for power distribution, grounding, and cooling systems[4][5]
•Conventional data center copper usage: 5,000–15,000 tonnes; AI-ready campuses require 3–4× that amount due to higher power loads and advanced liquid cooling with copper heat exchangers[3][5]
•Grid infrastructure demands: Hyperscale AI campuses require multiple redundant grid connections for uninterrupted power supply, with transmission lines, substations, and upgrades consuming large copper volumes[2]
•Data center infrastructure spending trajectory: North American investment rising from $33 billion (2020) to $70 billion (2030) and $185 billion (2040), with each AI-ready site locking in thousands of tonnes of copper for decades[2]
•Copper demand by sector through 2030: Electric vehicles use 80–100 kg per EV (vs. 20–30 kg for combustion vehicles); renewable energy and 5G infrastructure drive specialty metal demand[3]

🔮 Future ImplicationsAI analysis grounded in cited sources

Copper supply deficit will reach 10 million metric tons by 2040 unless mining capacity expands significantly.

Global production peaks in 2030 at 33 million metric tons then declines, while demand reaches 42 million metric tons by 2040, creating structural inelastic demand across data centers, EVs, and renewable grids with limited substitution options[1][4].

AI data center copper demand will drive price volatility through 2028, with peak consumption at 572,000 tonnes in 2028 before moderating.

BloombergNEF projects this peak mid-decade, followed by cumulative locked-in copper exceeding 4.3 million tonnes by 2035, creating a supply gap that could swell to 6 million tonnes by 2035[2].

Defense spending and emerging technologies (humanoid robots, AI infrastructure) will collectively add 4+ million metric tons of copper demand by 2040.

Defense demand is expected to triple by 2040, while 1 billion humanoid robots would require 1.6 million metric tons annually; these vectors represent new demand sources beyond traditional electrification[1].

⏳ Timeline

2026-01

S&P Global releases comprehensive copper supply-demand study projecting 10 million metric ton deficit by 2040 amid AI and defense spending surge[1]

2026-02

International Copper Study Group confirms 150,000 tonne refined copper shortfall for 2026, reversing previous surplus forecasts[3]

2028

AI data center copper demand projected to peak at 572,000 tonnes annually[2]

2030

Global copper production peaks at 33 million metric tons; data center consumption reaches 330,000–1.1 million tonnes annually (up to 3% of global demand)[1][3]

2035

Cumulative copper locked into data centers projected to surpass 4.3 million tonnes; supply gap could swell to 6 million tonnes[2]

2040

Global copper demand reaches 42 million metric tons (50% increase from current levels); supply deficit reaches 10 million metric tons unless mining capacity expands[1]