In her latest blog, Dr Ana Valdivia, Lecturer in AI, Government and Policy, Oxford Internet Institute, explores the structure of AI supply chains, the environmental impact of data centres on local communities, and why scholars and governments need to reconsider algorithmic harm through an environmental lens.
She sets out her arguments further in her new article ‘The Supply Chain of AI: the environmental impact of AI and the need to reconsider algorithmic harms’ just published in the journal Information Communication, and Society, calling for the entire production line of AI to be investigated from a critical and environmental lens.
The supply chain capitalism of AI
The supply chain of AI is a complex, global and opaque mechanism that extracts, produces and distributes components that are key for the making of AI nowadays. The supply chain of AI entangles other types of industries such as data centres, data labelling companies and big tech firms together with the electronic industry. Within this supply chain, there are different infrastructural elements that are key in the context of the making of AI. On one hand, mines, chips factories, data centres and e-waste dumps are key elements of the supply chain that manufacture products necessary for AI such as semiconductors, servers and wires, amongst others. On the other hand, the digital industry which comprises software companies and data labelling firms are also part of the AI supply chain. They design algorithms, create and curate datasets needed to fit AI systems.
However, digital elements such as datasets and algorithms together with human labour (data annotators, data scientists, data engineers, etc.) are also key elements in the supply chain of AI (see Figure 1). The making of AI entails different skills such as extracting minerals in mines, manufacturing semiconductors in factories, programming algorithms or curating datasets. These supply chains make visible the proliferation of workers in different industrial sectors that might belong to different classes and might be exposed to different mechanisms of exploitation and violence.
Figure 1. The supply chain capitalism of AI. This image partially captures the supply chain of AI as a global and complex phenomenon. Natural resources, components and materials to build AI infrastructure are extracted, shipped, manufactured and produced across the globe. For instance, NVIDIA obtains tungsten from Brazil; gold from Colombia and tantalum from Kazakhstan. Minerals are assembled to manufacture GPUs by TSMC. NVIDIA sells GPUs across data centres in the world. Given the refresh rates of these materials, data centres sent their components to recycle plants or dumps. The human labour wrapped-up in this chain includes, data labellers, logistics drivers, data scientists, miners, data centre operators and electronic waste dismantlers, who are also scattered across different geographies. Source: NVIDIA (Citation2022) and fieldwork.
The environmental cost of AI
Data centres play a crucial role within the AI supply chain, as they provide the vast amounts of computing power needed to train and run AI systems. With growing reliance on digital technologies, the number of data centres around the world has skyrocketed. In fact, the AI industry continues to drive up the number of data centres as it demands increasing amounts of processing power.
This comes with significant environmental and social costs. To meet the industry’s needs, there is more demand for land, energy and water, a trend that is expected to escalate in the coming years. Therefore, some data centre projects have sparked resistance from local communities opposed to the construction of data centres in their areas.
Water struggles within the supply chain capitalism of AI: The data centre hub in Queretaro, Mexico
Queretaro has become a major hub for data centres in Mexico, playing an important role in the growing AI supply chain. This state already hosts 10 data centres, with plans for 18 more to be built over the next decade.
Several factors have made Queretaro attractive to the data centre industry. The first relates to its industrial legacy. Queretaro is among the Mexican states experiencing rapid economic growth due to a large influx of foreign capital. In interviews, many participants explained that the region became a hub for many US automative factories in the 2000s, which transformed the economic and socio-demographic character of the region. After this boom, the aerospace industry expanded its presence in Queretaro. With this established industrial base makes it a prime location for data centre operations.
The second relates to its geographical situation. Queretaro is two hours’ drive from the capital, Ciudad de Mexico. Moreover, there are sea cables landing in Mexico which facilitates the faster transfer of data through the Internet.
A third factor that emerged through Dr. Valdivia’s fieldwork was the willingness of the local government to facilitate the installation of data centres within the region.
As the demand for generative AI grows, the supply chain supporting AI is expanding, and Queretaro is becoming a key location for this industry. As a consequence, the local government is enhancing the region’s critical infrastructure by improving the electrical grid and water supplies.
Figure 2. Electricity demand of incoming data centres in Querétaro (México). AI data centres (Microsoft) and online gaming (Origin) are going to use the most electricity. Source: Found during fieldwork (LX Legislatura de Querétaro, Citation2023).
Contested conflicts and water precarity in Queretaro
In 2006, the local government approached the community of Maconi with a proposal: they suggested extracting water from the area’s natural resources to supply the growing industrial parks in exchange for improving local infrastructure and access to drinking water. The community agreed. Subsequently, the government built one of the largest hydraulic infrastructures within Mexico: Acueducto II, (AQII) a 123km-long pipeline that pumps water from the mountains in Maconi to the citizens in Queretaro and its industrial parks.
However, seventeen years later, the local government not only failed to fulfil its promise to improve local infrastructure and improve access to drinking water, but the water resources of Maconi are now drying up. As Dr. Valdivia observed during her 3 months of fieldwork, the community of Maconi is now demanding that the local government prioritise their access to drinking water over the needs of industrial businesses.
Frictions within AI’s supply chain
Dr. Valdivia’s fieldwork in Mexico revealed the hidden side of AI’s supply chain, showing how it leads to resource exploitation, environmental harm and community resistance. AI’s rapid progress and associated infrastructure is commodifying nature in a territory where communities are suffering water scarcity, leading to environmental injustice and power inequality.
Conclusion
The case of Maconi shows that there are important dimensions of the AI supply chain that critical AI studies have often overlooked – specifically, the environmental harms and local resistance that arises when communities are caught in the crossfire of global supply chains.
Dr. Valdivia’s analysis and recent fieldwork in Mexico serve as a call to investigate the infrastructural impact of AI through an environmental and social justice lens.
Download the full paper, ‘The Supply Chain of AI: the environmental impact of AI and the need to reconsider algorithmic harms’ by Dr Ana Valdivia, Lecturer in AI, Government and Policy, Oxford Internet Institute, published in the journal Information Communication, and Society on October 30, 2024.
