In late 2025 and early 2026, communities in Arizona were mobilising against data center expansion, filing lawsuits, packing town halls, and pressuring state legislators to end tax incentives for facilities that would draw hundreds of millions of gallons a year from local water supplies. In Virginia, a community group that had already voted out town council members who supported a data center proposal was preparing for another fight. In Florida, residents mobilised against a facility the size of 260 football fields planned for St. Lucie County, as described in local reporting. Across the United States, opposition networks now span well over 180 activist groups (according to Data Center Watch’s tracking). Data Center Watch estimates $98 billion in projects were blocked or delayed in just one quarter of 2025.
This is what AI ethics looks like when it leaves the seminar room. Town halls. Courtrooms. Utility bills.
The Speculative Detour
Meanwhile, a parallel conversation is absorbing an extraordinary share of intellectual attention. The University of Sussex is hosting a symposium this July on AI consciousness and ethics. Anthropic has formally acknowledged the possibility that its language model may possess some form of moral status. Scholars across philosophy and computer science are engaged in sustained debate over whether large language models might have phenomenal experiences, whether we risk committing “mindcrimes” by mistreating them, and whether a precautionary principle demands we extend moral consideration to digital systems, just in case.
The allocation of intellectual resources here is worth scrutinising. As Cambridge philosopher Tom McClelland recently argued, while researchers devote increasing attention to the hypothetical suffering of AI systems, we already kill approximately half a trillion prawns every year, beings for which the evidence of sentience is substantially stronger than anything a language model has demonstrated. The priorities are revealing.
The problem is not that consciousness research exists. The problem is what it crowds out. When “AI ethics” becomes synonymous with speculative questions about machine sentience, the field loses sight of harms that are already measurable and already reshaping communities. Let us be direct: a significant portion of the philosophical discourse on AI functions as a distraction from the structural conditions under which these systems are built, powered, and imposed on populations who never consented to bear their costs. Often the philosophers involved are well intentioned. Sometimes they are genuinely rigorous. The effect is the same.
What the Material Record Shows
The empirical evidence is not ambiguous. The International Energy Agency projects that global data center electricity consumption will reach 945 terawatt-hours by 2030, more than doubling from 2024. In the United States, data centers already consume 4.4% of national electricity. That figure could triple to 12% by 2028. For much of the 2010s, efficiency gains moderated data-center electricity growth even as workloads soared, until the recent AI-driven surge. Then AI changed the equation: U.S- data-center electricity demand more than doubled by 2023, with AI servers a major driver.
Research from Harvard’s Electricity Law Initiative has found that the discounted electricity agreements between utilities and technology companies can directly raise consumer rates. In Virginia, residential bill increases have been sought in rate cases as grid investment accelerates; data-center load growth is widely cited as a major driver of new infrastructure needs. Global AI-related water withdrawal is projected to reach 4.2 to 6.6 billion cubic metres by 2027, equivalent to four to six times Denmark’s total annual water withdrawal. In The Dalles, Oregon, Google’s data centers already consume more than a quarter of the city’s water supply.
The communities responding to these realities are not suffering from a lack of “AI literacy,” even as policy discourse increasingly emphasises ‘AI literacy’. They are making normative judgments. They are evaluating whether the distribution of costs and benefits is just, whether the promises of economic development hold up against the evidence of rising bills and strained infrastructure, and whether their democratic institutions can protect them against corporations whose capital expenditure budgets exceed the GDP of small nations.
Three Conversations That Need Each Other
What we are witnessing is a fracture across three registers of analysis that ought to be integrated but rarely are.
The first is empirical and material: the measurable costs. Energy consumption, water diversion, carbon emissions, rising electricity bills. These are documented by agencies like the IEA, by investigative journalists, and by the communities themselves. They answer the question: what is actually happening?
The second is regulatory and institutional. The EU AI Act, which becomes fully applicable in August 2026, establishes risk classifications, transparency obligations, and fundamental rights protections. It is the most comprehensive regulatory effort to date. But scholarship has already identified a significant conflation at its core: the framework tends to treat “trustworthiness” as equivalent to “acceptability of risk,” collapsing a normative evaluation into a technical assessment. Whether a system meets a risk threshold is not the same question as whether the communities affected by that system consider it acceptable. Regulatory compliance is necessary. It is not sufficient.
The third is phenomenological and normative: the lived experience of people whose environments are transformed by technological infrastructure, and the evaluative frameworks through which they make sense of that transformation. This is the register that the consciousness debate occupies almost exclusively, aimed at the wrong subjects. The phenomenologically relevant question in 2026 is not whether a language model has inner experiences. It is what it means to live beside the physical substrate of “the cloud,” to watch your water table drop and to open an electricity bill that has increased by a third.
The analytical task is to trace the connections between these registers. The empirical data on energy consumption becomes meaningful only when situated within regulatory frameworks that determine who bears the cost. The regulatory frameworks become credible only when they engage the normative judgments of affected communities. And the normative dimension gains traction only when it is grounded in empirical reality rather than speculative scenarios.
Acceptance Is Not Acceptability
There is a distinction, fundamental to the study of technology’s social embedding, that this moment makes vivid: the difference between acceptance and acceptability. Acceptance is a behavioural fact. People use the technology, they adopt it, they integrate it into daily life. Acceptability is a normative judgment: the technology is good, its costs are justified, its governance is legitimate. The two do not necessarily coincide.
Millions of people use AI-powered tools every day. That is acceptance. Whether the infrastructure required to deliver those tools meets the standard of acceptability is an entirely different question, and one that cannot be answered by technologists or by philosophers speculating about machine consciousness. It can only be answered through genuine engagement with the people and institutions who bear the consequences.
Social acceptance is not something to be engineered through communication strategies after the fact. It is earned or forfeited by the material conditions a technology creates. Responsible innovation means ethics as a structuring condition of technological development, built in from the design stage through to deployment and operation, integrating empirical analysis, regulatory expertise, and normative evaluation throughout.
Where We Stand
At CyberEthicsLab,
this integration is the foundation of our work. Through European Innovation projects and consulting partnerships, we assess the ethics of artificial intelligence and the social acceptance of emerging technologies as dimensions of the same problem. We develop methodologies for ethics and privacy impact assessment, evaluate the social conditions under which technologies can be deployed responsibly, and support organisations in navigating the regulatory landscape, including the EU AI Act’s approaching full applicability.
The question that defines AI ethics in 2026 is not whether machines might one day deserve moral consideration. It is whether the people whose water is being diverted and whose electricity bills are rising are receiving moral consideration now. We work on the second question.
Note: This article was revised on 26 Feb 2026 to correct details regarding specific community actions cited in the opening paragraphs, to update statistical references for consistency with their primary sources, and to clarify technical terminology. The substantive analysis is unchanged.
