By Hyelyn Kim, Executive Director of the ARC Center
Artificial intelligence (AI) is often perceived as an advanced, immaterial industry that operates in a virtual space, driven primarily by intelligence and data. In reality, however, AI is a deeply material industry grounded in physical infrastructure. In particular, the rapid expansion of data centers has led to massive consumption of electricity and water, while their construction and operation require substantial material inputs. These facilities also drive the expansion of power infrastructure, including power plants, transmission networks, and batteries. And throughout all of these processes, large quantities of minerals are required.
The International Energy Agency (IEA) has analyzed that the proliferation of AI and the growth of data centers are significantly increasing demand for electricity infrastructure, which in turn drives rising demand for critical minerals. In practice, AI data centers are among the most mineral-intensive industries, relying on a wide range of metals such as copper, nickel, cobalt, and rare earth elements.
Even before the AI boom, the world had already been advancing carbon neutrality and energy transition centered on electrification. In 2021, the IEA presented a pathway for achieving net-zero emissions by 2050 through its report Net Zero by 2050. Around the same time, in The Role of Critical Minerals in Clean Energy Transitions, the agency made clear that this transition is inherently mineral-intensive and that mineral demand would surge under net-zero pathways.
According to the IEA’s Net Zero Emissions (NZE) scenario, by 2040, compared to 2024 levels, demand for lithium is projected to increase by approximately 7–8 times, graphite by more than threefold, nickel, cobalt, and rare earth elements by about twofold, and copper by around 1.5 times. With the addition of AI-driven industrial competition, mineral demand is expected to rise even further. The IEA estimates that the expansion of data center infrastructure alone could generate additional demand by 2030 of approximately 2% for copper, 2% for silicon, 3% for rare earth elements, and over 11% for gallium.
GPU semiconductors used in AI training are also highly metal-intensive. Recent studies show that one of NVIDIA’s widely used data center GPUs, the A100, is composed largely of metals such as copper, iron, silicon, and nickel. Training large-scale AI models requires thousands of such GPUs, resulting in the extraction of vast quantities of minerals—some of which ultimately become hazardous waste.
The expansion of mineral demand is directly linked to the expansion of mining, an industry with significant environmental and social impacts. Across all stages—from exploration and development to operation—mining causes deforestation, soil erosion, water pollution in rivers and oceans, and air pollution, all of which directly affect the health and livelihoods of local communities.
Mining is also closely associated with forced displacement, labor exploitation, community conflict, and violent disputes. A study analyzing environmental conflicts between 2011 and 2019 identified mining as the industry responsible for the largest number of such conflicts. Indigenous peoples and peasant communities were found to be both the most active in raising concerns and the primary groups affected. Other research indicates that more than half of critical mineral projects are located on or near the territories of Indigenous peoples and small-scale farming communities.
Indigenous and peasant communities have long lived in close relationship with nature and are internationally recognized as key stewards of ecosystems—often referred to as “guardians of nature.” Yet today, their lands are rapidly being degraded by large-scale extraction and industrial development, frequently justified under the banners of “energy transition” and “advanced industrial development.”
On Halmahera island in Indonesia, Indigenous communities with limited contact with the outside world still exist. Recently, large-scale nickel development projects led by Chinese capital have expanded in the region. Critics argue that land has been forcibly acquired without ensuring free, prior, and informed consent (FPIC), while large-scale deforestation has reduced hunting and gathering areas, threatening Indigenous livelihoods. Some global companies have raised concerns and initiated supply chain reviews, and certain European firms have withdrawn their investments. Nevertheless, the expansion of mining and smelting facilities in Halmahera continues.
South Korean companies are also investing across Indonesia in various stages of nickel production, including mining development, smelter operations, intermediate processing, and battery manufacturing. At these project sites, environmental pollution, labor rights violations, and conflicts with local communities continue to be reported by local media.
The South Korean government has also made securing critical minerals a key policy priority, promoting various forms of diplomatic and industrial cooperation with Indonesia, including memoranda of understanding (MOUs), and expanding support for related companies. However, institutional mechanisms to prevent, manage, and remedy human rights and environmental impacts in mining operations remain insufficient. These critical mineral supply chains are not only linked to electric vehicle batteries but are also deeply embedded in the core infrastructure of the AI industry, including data centers and semiconductors.
In this context, recent discourse surrounding the AI industry has been highly optimistic, and at times reflects narratives that strongly align with capital interests. Claims that AI will improve energy efficiency and reduce carbon emissions, thereby contributing to climate action, and arguments that industrial growth must be prioritized to secure geopolitical competitiveness are prominent. In this framing, constrained energy and mineral supplies are treated as “bottlenecks,” while opposition from local communities is often dismissed as “NIMBYism.”
However, such projections are largely based on limited and selectively disclosed data from corporations, and rarely account for the full environmental and social impacts across supply chains. In particular, there is little discussion of where and how mineral extraction—the material starting point of AI—takes place, and who bears the associated costs and damages.
In March, a subcommittee of the National Assembly’s Science, ICT, Broadcasting and Communications Committee passed the “Special Act on AI Data Centers.” The bill includes provisions such as deregulation of site selection, streamlined permitting procedures, and tax incentives. Notably, exemptions from power grid impact assessments in non-metropolitan areas and special provisions for power purchase agreements (PPAs) have emerged as key points of contention.
A power grid impact assessment is a system that evaluates whether existing power networks can accommodate new large-scale electricity facilities. It serves as a minimum safeguard to ensure stable operation and balance in the power system by assessing risks such as transmission overload, voltage instability, and overall grid reliability. Granting blanket exemptions from such assessments to an entire industry sector represents a highly unusual form of regulatory relaxation.
PPAs, which are long-term contracts between electricity producers and consumers, have in practice been used primarily as mechanisms for procuring renewable energy, particularly solar and wind. Nevertheless, the inclusion of special PPA provisions in the bill has raised concerns that it may pave the way for expanded use of other energy sources, such as LNG or small modular reactors (SMRs).
In response, civil society organizations in South Korea have called for a halt to the bill’s deliberation. They argue for institutional measures such as disclosure of data center electricity and water consumption, preparation of environmental safeguards, addressing regional imbalances in electricity supply, expanding renewable energy use, and evaluating the efficiency of energy infrastructure.
These concerns highlight the significant energy burden and local impacts associated with data center expansion. However, considering that the material starting point of AI lies in mineral extraction—and that South Korean capital and policy are deeply involved in these supply chains—such discussions must extend beyond electricity issues to encompass mineral supply chains.
The expansion of the AI industry generates new environmental and social impacts along mineral supply chains, with the most severe consequences often occurring at the upstream end—far removed from public attention. Therefore, institutional frameworks must address human rights and environmental impacts across the entire mineral supply chain. At the same time, efforts to reduce demand for new extraction—such as strengthening mineral recycling and advancing a circular economy—must be pursued in parallel. Companies must also transparently disclose cases of human rights abuses and environmental damage within their supply chains, along with the measures taken in response.
The development of the AI industry must move beyond the illusion of immaterial innovation and take responsibility for the material impacts that underpin it. Closing this gap requires strengthening institutional frameworks to ensure accountability across the full chain of production.
This article was originally published in Pressian on April 9, 2026.