AI Touches Down on Biodiverse Ground

By Javier Surasky

Artificial intelligence lives in models, data, computing power, and human talent, and it has profound effects on societies and their dynamics. Yet this entire chain rests on a common foundation: the natural resources that sustain chip manufacturing, the construction and operation of data centers, electricity grids, and the other infrastructure required to develop and use artificial intelligence systems.

Before becoming technology, AI was also Pachamama. This may seem fairly obvious, but it has a direct consequence that receives far less attention: decisions and public policies concerning the use of finite natural resources are also decisions about AI’s moment zero.

Kate Crawford (2021, p. 8) captures this through a deliberately provocative formulation when she argues that, within the expression artificial intelligence, the word “artificial” obscures digital technology’s dependence on raw materials, energy, human labor, infrastructure, and logistics.

Since Crawford wrote those words, the material dimension of AI has become even more extreme. In 2025, global electricity consumption by data centers grew by 17%, while consumption by centers specifically oriented toward AI increased by 50%. In just eighteen months, the capacity of frontier data centers specifically designed for AI tripled (International Energy Agency [IEA], 2026a).

Speed matters because every new piece of infrastructure requires electricity, water, land, networks, equipment, and minerals, while also creating commitments and dependencies that may later prove costly to change.

For that reason, I believe debates about AI need to expand to give proper weight to the territories that sustain its infrastructure, the resources committed to its development, and the ways in which environmental costs are distributed across societies and generations.

The underlying question is easy to formulate and much harder to answer: what use of our resources and ecosystems are we willing to accept in order to build the artificial intelligence we want?

I will approach this question through international environmental law, focusing on three of its principles: permanent sovereignty over natural resources, sustainable development linked to intergenerational equity, and precaution.

Sovereignty Does Not Mean Unlimited Availability

The principle of permanent sovereignty over natural resources emerged in connection with decolonization, self-determination, and struggles over economic control of the resources of newly independent countries. It was a legal and political response to a history in which colonial resources had been treated as material reserves serving projects decided by colonial powers.

The principle recognizes the sovereign right of States to exploit their own resources in accordance with their environmental and development policies, while peoples hold title to those resources. Neither form of authority is exercised in a vacuum, nor can either be understood as a blank check to make decisions about natural resources without considering their social, environmental, and intergenerational consequences.

Anghie allows us to read this issue from a critical historical perspective: for postcolonial States, formal sovereignty did not guarantee real economic decision-making capacity, and struggles over sovereignty and resources were also struggles over the material conditions needed to sustain newly won independence and over the room for maneuver available within international economic structures built, to a considerable extent, under conditions of inequality (Anghie, 2004).

The digital economy brings that debate back into view. AI promises efficiency, innovation, investment, and competitiveness, but its deployment depends on resources located somewhere: energy, water, land, minerals, and infrastructure. Asking who decides how those resources are used and who captures the benefits is therefore central to understanding the political workings of AI.

Water offers a useful example. An IEA report published in October of this year estimates that nearly 40% of identified AI-dedicated data center capacity is located in areas facing high or extremely high baseline water stress (IEA, 2026b).

At the same time, research published in 2025 by Lawrence Berkeley National Laboratory found differences of more than ten thousandfold in the water consumption associated with different data center workloads, depending on technological and environmental factors (Lei et al., 2025). It therefore makes little sense to speak of a universal figure for how much water “AI consumes”: where we build the infrastructure and what kind of infrastructure we build become critical questions.

Minerals pose a similar problem. Advanced digital systems depend on critical minerals used in semiconductors, servers, and specialized equipment. The IEA (2026c) shows that advances in digital and related technologies are increasing the strategic importance of minerals such as gallium and germanium, whose production and refining are highly geographically concentrated.

The expansion of this demand has uneven consequences. The International Resource Panel warns that expanding extraction without changing current practices could intensify social and ecological impacts in resource-producing territories (United Nations Environment Programme & International Resource Panel [UNEP & IRP], 2025).

I propose reading these processes as part of an obscured geography of AI, not simply an invisible one, since making it invisible requires work. This geography casts a veil over the way benefits become concentrated where systems are developed and consumed, while part of their environmental costs are displaced onto the territories where materials are extracted, energy is produced, or the infrastructure needed to sustain those systems is installed.

From this perspective, sovereignty over natural resources becomes a condition of digital transformation. If decisions about land, water, energy, minerals, and ecosystems are left outside the technological debate, we also exclude a substantial part of our democratic capacity to decide about the future.

Innovation Is Not the Same as Development

The second principle is sustainable development, classically defined by the Brundtland Commission as development that meets present needs without compromising the ability of future generations to meet their own (World Commission on Environment and Development [WCED], 1987).

That definition requires an important distinction: innovation and development may coincide, but they are not equivalent. A technology can be sophisticated, profitable, and globally competitive while also compromising finite natural resources, reproducing inequalities, narrowing the options available to those who will come after us, and affecting today’s vulnerable groups more severely.

Keith Skene approaches this question through an unsettling one: if human intelligence has been part of the environmental crisis, why should an AI built within the same economic and social systems automatically lead us toward a sustainable transition? His answer is cautious: AI may contribute to that transition, but “not as it stands at present” (Skene, 2020, p. vii).

AI can provide tools for environmental monitoring, climate modeling, risk management, energy efficiency, precision agriculture, and ecosystem observation. These are important uses, but on their own they tell us little about the sustainability of the material system required to produce and expand those technologies.

Biodiversity allows me to take this question one step further.

In October 2026, COP17 of the Convention on Biological Diversity, held in Yerevan, conducted the first global review of progress in implementing the Kunming-Montreal Global Biodiversity Framework. The State of Biodiversity Action concluded that the world is not on track to achieve the 2030 biodiversity targets adopted only four years ago (Secretariat of the Convention on Biological Diversity [CBD], 2026a).

COP17’s political response was the Yerevan Pathway, which identifies priorities for accelerating action through 2030, including reducing subsidies and incentives harmful to nature and integrating biodiversity into national budgets and economic decision-making (CBD, 2026b).

Read in light of the problem I am raising here, that agenda forces us to look toward AI as well. If economic decisions are expected to incorporate biodiversity, it is difficult to justify excluding from that calculation policies designed to attract data centers, expand energy infrastructure, secure mineral supplies, or develop national AI ecosystems.

A technology policy that counts investment, computing capacity, and employment while externalizing environmental costs offers an incomplete picture of its results. Here we encounter a connection that I think is particularly important: digital governance asks who controls models, data, infrastructure, and computing power, while environmental governance asks how ecosystems should be preserved, resources used sustainably, and responsibilities and benefits distributed. The two need to work coherently and in coordination.

At the same time, COP17 identified limited access to technology, data, and knowledge among the obstacles to implementing the Global Biodiversity Framework. Technology can therefore help protect nature while simultaneously increasing pressure on it. This ambivalence makes it insufficient to classify AI as environmentally positive or negative. The question is under what conditions it is developed, where its costs fall, and who retains the capacity to decide about them.

Skene argues that sustainable thinking has ceased to be an ethical supplement and has become a structural necessity. AI must therefore also be assessed in terms of its compatibility with an ecological transition that is both just and respectful of biodiversity.

Intergenerational equity reinforces the point: land, water, energy, minerals, and ecosystems are material conditions of future freedom, and a generation that irreversibly compromises them reduces the political room for maneuver available to those that follow.

Precaution: Deciding Before the Damage Becomes Irreversible

The third principle is precaution. Its main value lies in providing a rule of prudence in the face of serious or uncertain risks: where scientific uncertainty exists alongside the possibility of serious or irreversible harm, the absence of absolute certainty cannot automatically become a justification for postponing preventive measures. Uncertainty should not serve as permission to move ahead without assessment, public information, deliberation, and safeguards.

Barboza provides a useful basis for thinking about this problem through international environmental law: prevention in relation to risky activities includes risk assessment, information, notification, consultation, and the balancing of interests (Barboza, 2011).

This matters when we confront technologies presented as inevitable, because technological inevitability tends to depoliticize decisions that depend on permits, investment, electricity grids, public infrastructure, and land use.

And where there is a risk of serious or irreversible harm, deciding too late is also a decision.

From this diagnosis, I believe precautionary reasoning must also take into account the speed at which difficult-to-reverse decisions can become entrenched, because once infrastructure is installed and contractual commitments concerning energy and access to other natural resources have already been made, the options available for subsequent action become narrower.

The OECD’s 2025 revision of its Recommendation on Digital Technologies and the Environment adopts an approach that calls for considering the environmental footprint of these technologies throughout their entire material existence, improving the measurement and transparency of their impacts, and strengthening coordination among digital, environmental, and energy policies (Organisation for Economic Co-operation and Development [OECD], 2025).

Precaution also shifts the burden of justification: those promoting high-impact territorial transformations should justify their necessity, disclose their costs, make their risks transparent, and assess alternatives.

The Cloud Needs Land Too

For a long time, the metaphor of “the cloud” encouraged us to imagine digital infrastructure as almost immaterial. Today we know that this supposed immateriality has very earthly expressions in hyperscale data centers with high energy and water consumption, submarine cables, and chip production, all of which intensify demand for finite minerals extracted from somewhere.

Digital chains can therefore draw their own geography and be “mapped” onto territories that sustain life and the conditions for its reproduction. Sovereignty over natural resources must also be exercised as an intergenerational responsibility. Ultimately, there can be no technological sovereignty without ecological sovereignty, and no digital justice without climate justice.

The promise of democratizing AI encompasses decisions about algorithms, territories, natural resources, and the location of infrastructure. My reading of the evidence gathered here is that AI governance must extend to this reality, interacting with existing international spaces for environmental protection and assuming responsibilities of its own.

The current pace of AI expansion shows that its growth is moving at a rate that electricity grids, supply chains, and planning processes are struggling to keep up with (IEA, 2026a), implying a pattern of use that is unsustainable even in the short term.

References

Anghie, A. (2004). Imperialism, sovereignty and the making of international law. Cambridge University Press.

Barboza, J. (2011). The environment, risk and liability in international law. Martinus Nijhoff Publishers.

Crawford, K. (2021). Atlas of AI: Power, politics, and the planetary costs of artificial intelligence. Yale University Press.

International Energy Agency. (2026a). Key questions on energy and AI. IEA.

International Energy Agency. (2026b). Powering artificial intelligence: Strategies for integrating data centres into electricity systems. IEA.

International Energy Agency. (2026c). Global critical minerals outlook 2026. IEA.

Lei, N., Lu, J., Shehabi, A., & Masanet, E. R. (2025). The water use of data center workloads: A review and assessment of key determinants. Resources, Conservation and Recycling, 219, 108310.

Organisation for Economic Co-operation and Development. (2025). Revised recommendation of the Council on digital technologies and the environment (C/MIN(2025)10/FINAL). OECD.

Secretariat of the Convention on Biological Diversity. (2026a). State of biodiversity action: The global report on collective progress. Convention on Biological Diversity.

Secretariat of the Convention on Biological Diversity. (2026b). UN Biodiversity Conference charts Yerevan Pathway to accelerate global action for nature. Convention on Biological Diversity.

Skene, K. R. (2020). Artificial intelligence and the environmental crisis: Can technology really save the world? CRC Press.

United Nations Environment Programme & International Resource Panel. (2025). Financing the responsible supply of energy transition minerals for sustainable development. UNEP.

World Commission on Environment and Development. (1987). Our common future. Oxford University Press.