AI Runs on Electricity: Also on Cobalt, Copper, and Geopolitics
Artificial intelligence is often described in abstract terms: algorithms, models, and data. Yet behind every AI data center sits a vast physical infrastructure built from mined minerals, manufactured chips, and globally distributed supply chains. As AI data centers expand rapidly, sustainability discussions are shifting from electricity and water toward a deeper question: what material resources make this digital ecosystem possible, and how resilient and responsible are those supply chains?
At the core of modern AI infrastructure is specialized hardware, especially graphics processing units. AI training clusters rely on dense arrays of GPUs that contain critical minerals such as copper, cobalt, nickel, lithium, tantalum, palladium, and rare earth elements. Each material carries environmental and geopolitical implications. Mining can involve deforestation, water contamination, and emissions, while processing is often concentrated in a small number of countries. As demand for AI accelerators rises, pressure on these resource streams increases.
Advanced semiconductor fabrication requires ultra-pure silicon, specialty gases, and rare metals deposited in microscopic layers. Producing a single high-end chip involves hundreds of steps across a globally distributed supply chain spanning dozens of countries. The AI boom is therefore not only a software revolution but also a manufacturing expansion, tying digital progress to physical extraction and industrial capacity. Analysts increasingly describe AI infrastructure as a new category of strategic resource consumption comparable to earlier industrial expansions.
Geopolitics plays a central role. Many minerals used in semiconductors are geographically concentrated. Cobalt production is closely tied to the Democratic Republic of Congo. Rare earth processing is dominated by China. Lithium refining is similarly concentrated. Leading-edge chip fabrication depends on facilities in Taiwan, South Korea, and a handful of regions. This concentration introduces supply risk because political tensions, trade restrictions, or regional instability can ripple through the AI hardware ecosystem.
JUNIOR KANNAH / AFP via Getty Images
Recent policy shifts show how quickly these dynamics can change. Export controls on advanced chips, restrictions on manufacturing equipment, and incentives for domestic fabrication have become tools of economic strategy. Governments increasingly view advanced computing infrastructure as a critical national capability. AI data centers now sit at the intersection of sustainability, industrial policy, and national security, and the environmental footprint of hardware production cannot be separated from where materials originate and components are manufactured.
This creates a sustainability dilemma. Scaling AI requires more chips, servers, and raw materials, yet expanding mining and processing intensifies environmental pressures. Life cycle analyses show that a substantial share of a server’s footprint occurs before it is powered on during extraction, refining, and manufacturing. Ignoring this embedded impact risks understating the true resource cost of digital infrastructure.
Circularity offers one of the most promising responses. Many metals used in electronics can be recovered and reused. Recycling servers, batteries, and circuit boards can reclaim copper, aluminum, gold, and rare metals that would otherwise require new mining. Some companies design hardware with modular components so parts can be replaced without discarding entire systems, while others invest in urban mining initiatives that extract materials from retired electronics. These approaches shift the model from linear consumption toward a loop where materials remain in circulation longer.
Circular strategies provide environmental and operational benefits. Recovered materials reduce exposure to supply disruptions and price volatility. Secondary markets can lower costs, and extending hardware lifespans reduces both capital expenditure and waste. Circularity, therefore, aligns resource efficiency with resilience, making infrastructure less dependent on fragile global supply chains.
Still, circularity alone will not solve the resource challenge. Recycling rates for many critical minerals remain low because components are difficult to disassemble or contain only small quantities of valuable elements. Improving recovery requires better product design, standardized recycling, and policy incentives that make reclamation viable. Transparency is also essential. Companies often report energy and carbon metrics in detail, yet disclosures about material sourcing, recycled content, or hardware life cycle impacts are far less consistent.
What emerges is a broader understanding of AI infrastructure. It is not just a network of servers but a material system rooted in geology, chemistry, logistics, and geopolitics. The sustainability of AI data centers, therefore, depends not only on renewable electricity or efficient cooling but also on how responsibly the industry sources minerals, manufactures hardware, and manages end-of-life equipment. The long term trajectory of artificial intelligence will hinge on whether its physical foundations can evolve as rapidly as its software capabilities, and whether those resources can be managed in a way that sustains both technological progress and the planet that supports it.
