Artificial intelligence is often described as intangible: weightless code, abstract models, and invisible intelligence. This is a convenient fiction. In reality, AI is one of the most material technologies ever developed. It runs on electricity and consumes water at industrial scale. It depends on vast mineral supply chains and occupies physical space in ways that dwarf previous digital technology.
The illusion of immateriality has allowed policymakers—especially outside Europe and across the Atlantic—to delay confronting its real-world consequences. But those consequences are now impossible to ignore. AI is rapidly becoming one of the most energy-intensive, resource-dependent, and politically consequential infrastructures of the twenty-first century. And yet, governance remains fragmented, insufficient, or altogether absent.
In this emerging landscape, one actor stands apart. The European Union—often criticized for moving slowly, regulating early, and prioritizing safeguards—may in fact be the only actor capable of governing AI at the scale and seriousness the moment demands. In a world of technological acceleration and geopolitical rivalry, Europe is, quite simply, the only adult in the room.
AI is Infrastructure, Not Software
The first mistake in the global AI debate is conceptual. AI is still treated as a digital service rather than what it has become: a foundational infrastructure layer. This distinction matters because infrastructure is governed differently. It shapes economies, redistributes power, and creates long-term dependencies. It requires planning, regulation, and coordination across sectors and borders. AI already exhibits all these characteristics.
Data centers—the backbone of AI—consume vast amounts of electricity. According to the 2025 IEA “Energy and AI” report, in 2024 alone global data center demand reached approximately 415 terawatt-hours, roughly equivalent to the annual electricity consumption of Japan. By 2030, this figure could more than double to 945 terawatt-hours, driven primarily by AI workloads.
That is two Japans.
At that scale, AI is no longer a marginal addition to the digital economy. It is a structural component of national energy systems. It competes with households and industry for electricity. It influences grid planning, investment decisions, and emissions trajectories. And unlike traditional infrastructure, its current growth is exponential. A single large AI data center can consume as much power as 100,000 homes. The largest facilities now under construction may approach the electricity demand of Paris. This is not a future scenario. It is already underway.
The False, Dangerous Comfort of “Green AI”
There is a growing narrative, particularly among technology companies, that AI will ultimately be a force for sustainability—that it will optimize energy systems, reduce waste, improve efficiency, and accelerate climate solutions. There is truth in this. AI can and does support grid optimization, improve renewable energy integration, detect methane leaks, and enhance industrial efficiency. Under the right conditions, it can contribute meaningfully to emissions reduction.
But this narrative is incomplete, and dangerously so. It conflates two distinct concepts: AI for sustainability and sustainable AI. The former refers to using AI to improve environmental outcomes in other sectors. The latter concerns the environmental footprint of AI itself: its energy use, water consumption, hardware requirements, and waste generation. These are not automatically aligned. A government can heavily incentivize investment in AI applications for solving climate change while simultaneously expanding energy-intensive data center infrastructure powered by fossil fuels. A company can develop AI tools for sustainability while obscuring the environmental cost of training and running its models.
Without comprehensive governance, the benefits of AI risk being outweighed—or at least diluted—by its direct environmental impact. This is further compounded by a well-known economic dynamic: the rebound effect. Efficiency gains reduce costs, which in turn stimulate demand. In the context of AI, more efficient models can lead to more widespread use, offsetting or even exceeding the environmental savings they generate. Efficiency without governance does not guarantee sustainability. It may simply accelerate consumption.
The assumption that markets will naturally align AI growth with sustainability is flawed. Renewable energy, while expanding rapidly, cannot currently meet the pace and scale of AI-driven demand. Today, renewables account for only a small but growing share of data center electricity, and even by 2030 they are expected to meet only about half of the additional demand generated.
The reasons are structural. Today’s grid infrastructure is not designed to integrate renewable capacity quickly enough. AI workloads require constant, uninterrupted power, which intermittent sources such as wind and solar cannot reliably provide without significant storage solutions. Likewise, the timelines are fundamentally mismatched: AI infrastructure can be deployed in months, while energy infrastructure takes years to build.
The result is predictable. In the absence of intervention, AI expansion is actively driving new fossil fuel generation, with serious environmental consequences. This is not a theoretical concern. It is a systemic risk—and precisely the kind of challenge that markets, focused on speed, scale, and short-term returns, are ill-equipped to address.
The Hidden Supply Chain
Energy, however, is only the beginning.
Beyond electricity, AI depends on a complex and largely overlooked physical supply chain whose geography is brutally concrete and deeply geopolitical. The cobalt and rare earth elements that underpin digital systems are extracted in the Democratic Republic of the Congo, often at significant environmental and human cost—what is frequently and euphemistically described as “artisanal mining.” China is home to a large share of these natural resources and dominates the refining and processing of many critical materials. This gives Beijing a structural position of leverage over the upstream inputs of the AI economy—increasingly recognized by other powers as a strategic vulnerability.
Further along the chain, the most advanced semiconductors are manufactured in Taiwan, making the island not merely an industrial hub but a geopolitical fault line. The concentration of cutting-edge chip production in a single, highly contested territory introduces systemic risk into the global AI ecosystem. Any disruption, whether political, military, or economic, would reverberate across every sector dependent on advanced computation. Europe, however, is not absent from this map. In the Netherlands, ASML occupies a unique strategic position: its extreme ultraviolet lithography machines are indispensable to the production of the world’s most advanced chips, giving Europe one of the few genuine industrial bargaining chips in the global semiconductor order.
At the far end of the chain, the afterlife of AI hardware tells a different story. Countries such as Nigeria and Ghana have become destinations for the world’s electronic waste, where discarded servers, devices, and components are dismantled under appalling conditions—described, again euphemistically, as “informal”—often through open burning and toxic exposure. Places like Agbogbloshie in Accra—once described as the world’s largest e-waste site—have come to symbolize this reality: a global digital economy sustained not only by innovation, but by the systematic externalization of environmental and human costs.
This is what AI supply chains look like: minerals from Congo and China, refined through Chinese-controlled processes, chips produced in geopolitically contested Taiwan, and waste accumulating in West Africa. A system of global connectivity, built on infrastructures of profound asymmetry, where power, profit, and environmental burden are distributed along sharply unequal lines—and where deepening dependencies offer little recourse to those bearing the greatest costs.
A Fragmented World
The global governance of AI reflects a deeper divide in political models. The United States presents itself as the champion of innovation. In practice, however, its AI ecosystem is increasingly shaped by a small number of dominant firms—private actors with enormous influence over infrastructure, standards, and direction. This is not a neutral market. It is a form of technological oligarchy. A handful of companies control the core layers of AI development: cloud infrastructure, model training, and deployment platforms. Their incentives are clear—scale, speed, and market dominance—and their accountability is limited. The result is a system driven not by public interest but by private optimization.
This gap between technological power and societal responsibility is increasingly visible, and increasingly recognized by the American public itself. The Bentley University-Gallup “Business in Society Report 2025” notes that approximately 69 percent of American have no or not much trust in businesses using AI responsibly. This is a striking figure for a country that positions itself as the global leader in technological innovation, and one that reveals a deeper contradiction: the system producing AI is not trusted even by those it is meant to serve.
China offers a different model: state-led coordination and strategic control, capable of mobilizing resources at scale, aligning AI with industrial policy, and integrating it into national planning. But this model is not designed for global trust. Its lack of transparency and centralized control limit its capacity to serve as a foundation for internationally legitimate standards.
Between oligarchy and state control, neither model provides a credible pathway for globally legitimate governance.
One of the most striking features of AI’s environmental impact is how little we actually know about it. There is no standardized methodology for measuring AI energy use. Companies treat key data—model architecture, data center location, and per-query energy consumption—as proprietary trade secrets. Public figures are inconsistent, often misleading, and rarely comparable across organizations.
Even within the European Union, where reporting requirements are advancing, significant gaps remain. The Energy Efficiency Directive mandates for data centers based in the EU disclosure. The EU AI Act requires reporting of training energy for large models, applicable globally. These are unprecedented steps, and they have provoked strong reactions not only from concerned companies but from the United States as well. Without shared metrics, however, there can be no meaningful comparison, no accountability, and no basis for international cooperation. Standard-setting, once again, is not merely technical exercise. It is geopolitical one.
Europe’s Moment
This is where Europe’s distinct approach becomes decisive. The European Union has built its identity not on technological dominance but on governance capacity. It regulates markets, defines standards, and aligns economic activity with societal objectives. It is a political and economic union in which Member States voluntarily pool sovereignty to find common solutions and agree on joint regulation applying across to all 27 EU Member States. As in many other policy domains, this is not a weakness. It is a strategic advantage.
The EU AI Act is poised to become the global gold standard for AI governance. Its regulatory approach is grounded in European values, with sustainability at its core. The Act introduces the first binding requirements for energy transparency in general-purpose AI models. The Energy Efficiency Directive mandates reporting on data center energy and water use across the EU. The Sustainability Rating Scheme establishes a framework for evaluating environmental performance. These are not isolated policies. They form the foundation of a governance model that treats AI as infrastructure—something that must be measured, constrained, and aligned with public goals.
Crucially, Europe’s regulatory reach extends beyond its borders. Any AI system deployed within the EU market is subject to its rules, regardless of where it was developed. This is how global standards are set: not through dominance, but through access. The same dynamic that made GDPR a global benchmark is now unfolding in AI governance. The Brussels Effect, in action.
AI is inherently global. Its development depends on cross-border data flows, distributed infrastructure, and interconnected supply chains. Climate change, too, is global, its causes and consequences indifferent to national boundaries. Connectivity is therefore not optional. It is the defining condition of both AI and sustainability. But connectivity without governance is unstable.
When systems are globally interconnected but locally regulated—or not regulated at all—costs are externalized and environmental impacts shifted across borders. Accountability becomes diffuse. A data center built in one jurisdiction can draw on energy systems, water resources, and supply chains extending far beyond it. An AI model deployed globally can generate environmental costs that no single authority measures or manages. This mismatch between global systems and fragmented governance is one of the central challenges of our time. Europe’s approach offers a way forward.
Sovereignty in a Connected World
For Europe, governing AI is not only about regulation. It is about sovereignty—and this does not mean isolation. On the contrary, it means the capacity to make independent strategic choices within an interconnected system.
Today, Europe remains structurally dependent on external providers, particularly U.S. hyperscalers—the dominant cloud infrastructure providers such as Amazon, Microsoft, and Google—for cloud infrastructure and AI capabilities. This dependence limits its ability to fully enforce its own standards. At the same time, Europe’s relationship with China must be defined on its own terms, not as an extension of U.S. strategic priorities. A credible governance model requires autonomy: technological, regulatory, and political. This includes investment in European cloud infrastructure, semiconductor capacity, and AI development ecosystems—not to compete for dominance, but to ensure that governance is not undermined by dependence.
Reducing that dependence will not be a purely technical transition. It will come with real political cost. Washington has long encouraged European allies to invest more in their own capabilities, yet when that independence begins to materialize, the reaction is far less supportive. Recent tensions illustrate the contradiction: when major Danish pension funds reduced their exposure to U.S. assets, the response from across the Atlantic was not indifference but pressure.
That same dynamic is likely to intensify sharply in the digital domain. As long as Europe relies on American hyperscalers, the transatlantic relationship remains structurally asymmetric. But a credible European cloud ecosystem—capable of hosting its own AI systems and governed under its own regulatory framework—would shift that balance. Such a move would not go unanswered. It would challenge not only commercial interests but the broader architecture of U.S. technological dominance.
This moment of friction also signals something larger: the early formation of an alliance of middle powers—Europe, Canada, and others—seeking to navigate between U.S. oligarchic innovation and Chinese state control. These are actors that share a commitment to open markets, democratic accountability, and rules-based governance, yet find themselves increasingly squeezed between two competing models of technological power. AI infrastructure may well be the first domain in which this coalition asserts itself—not as a bloc of confrontation, but as a bloc of governance, committed to connectivity and determined to shape its own terms.
Governing the System We have built
The rise of AI presents a paradox. It offers tools that could help address climate change, improve efficiency, and support sustainable development. At the same time, it risks intensifying resource consumption, energy demand, and global inequality. Resolving this paradox is not a technical challenge. It is a political one—and it requires recognizing AI for what it is: infrastructure that must be governed, not merely developed.
In a world where major powers are driven by the paradigm of control, Europe’s emphasis on rules, standards, and accountability may appear cautious. In reality, it is indispensable. Europe is not perfect. But in this moment, it is the only actor attempting to align technological progress with environmental sustainability and democratic legitimacy. That is what it means to be the only adult in the room. And it is why the future of AI’s environmental impact may ultimately be decided not in Silicon Valley or Beijing—but in Brussels.
Without governance, connectivity becomes chaos. Without standards, innovation becomes extraction. Without accountability, power goes unchecked.
