Dark side of AI supply chains: Exploitation, secrecy, and e-waste

Artificial intelligence has become one of the fastest-growing technologies in the world, but researchers are sounding the alarm about the hidden supply chains that power it. A new study published in the journal AI & Society argues that the expansion of generative AI rests on opaque and extractive networks with major political, social, and environmental consequences.

The paper, titled “The politics of artificial intelligence supply chains”, examines how AI systems are sustained by complex global infrastructures and identifies the political struggles embedded in their development. Using OpenAI’s ChatGPT as a case study, the authors outline how supply chains span mineral extraction, semiconductor manufacturing, data centers, digital labor, model development, cloud infrastructure, and ultimately the management of e-waste. They argue that far from being neutral, these chains are sites of power struggles, ecological harm, and resistance.

How are AI supply chains structured?

The researchers introduce a framework that breaks down AI supply chains into four interconnected spheres: AI infrastructure, AI preparation, AI deployment, and e-waste. This framing makes visible the hidden processes behind the sleek interfaces of popular AI systems.

AI infrastructure includes the extraction of minerals like lithium and cobalt, essential for powering chips and data centers. These resources often come from regions where extraction is linked to environmental degradation and the exploitation of local communities. The infrastructure layer also involves chip production, dominated by a small number of firms such as NVIDIA, TSMC, and ASML. This concentration of capacity creates strategic bottlenecks that are critical to global competition.

AI preparation encompasses the massive datasets required to train generative models. This process often relies on human data work, such as labeling, filtering, and moderating content. Workers, frequently based in lower-income regions, are tasked with repetitive and psychologically taxing jobs for low wages, forming an invisible labor force behind machine learning.

AI deployment refers to the software ecosystems, cloud infrastructure, and corporate platforms that deliver AI services worldwide. Tech giants such as Amazon, Microsoft, and Google dominate this stage, embedding AI into global commerce and communication networks. Finally, e-waste highlights the lifecycle of AI technologies, drawing attention to the disposal of servers, chips, and other hardware after their short operational lives, raising questions about sustainability.

By tracing these four spheres, the study underscores that AI supply chains are not merely technical systems but deeply political networks spanning multiple geographies and industries.

What political issues do these supply chains raise?

The authors identify three central political challenges in today’s AI supply chains: opacity, concentration of power, and coalitional politics.

Opacity is one of the defining features of AI production. Corporate secrecy, competitive pressures, and geopolitical sensitivities conceal how AI systems are built, making it difficult for regulators, civil society, or the public to assess their true costs. The hidden nature of data sourcing, labor practices, and environmental impacts means that many of the harms tied to AI development remain invisible.

Concentration of power is another major concern. A handful of companies dominate key areas, from chip production and cloud services to advanced model development. This concentration creates global dependencies, limiting competition and giving disproportionate influence to a small set of actors. The study highlights how these bottlenecks are central to ongoing geopolitical tensions, particularly between the United States and China, as each seeks to secure supply chain resilience and technological leadership.

The third issue, coalitional politics, captures the push and pull between expansion and resistance. On one side are alliances of corporations and governments promoting the growth of AI infrastructure as a driver of economic competitiveness. On the other side are coalitions of workers, indigenous communities, environmental activists, and artists resisting exploitative practices. These groups highlight labor rights violations, ecological harm, and the use of creative works without consent in training datasets. The study notes that these emerging resistances challenge the narrative of inevitable AI progress by demanding accountability and fairer governance.

Why does this matter for the future of AI?

The research asserts that AI supply chains should not be seen as neutral pipelines of innovation but as contested political terrains. Each stage of the chain involves power struggles, from resource extraction and labor conditions to data governance and e-waste management.

The authors argue that addressing these issues requires stronger regulatory oversight, more transparency, and recognition of the human and ecological costs embedded in AI systems. Without intervention, the risks of exploitation, environmental damage, and monopolization will continue to grow alongside the rapid expansion of generative AI.

Notably, the study also points to the potential for coalitional resistance to shape the future trajectory of AI. Worker movements, indigenous activism, environmental campaigns, and advocacy from creative communities have already influenced debates around labor conditions, ethical data use, and sustainability. These forms of opposition signal that AI’s future is not predetermined but open to political contestation.

By mapping out AI supply chains as political networks, the research provides a framework for policymakers, activists, and scholars to better understand where interventions are needed. The authors stress that meaningful governance must go beyond regulating algorithms and address the entire material and labor-intensive infrastructures that make AI possible.