There is a moment in the life of every large technology when it stops being a matter for engineers and becomes a matter for town councils. Genetically modified crops reached that moment in the late 1990s. AI data centres reached it somewhere around the middle of last year.
The comparison is now being made often enough that it deserves proper examination rather than casual assertion. Are the campaigns against hyperscale data centres in Virginia, Tucson, Dublin, Santiago and Cape Town a rerun of the anti-GMO movement, or is something structurally different going on? The honest answer is that the parallels are real and instructive, but they break in places that matter a great deal for how this ends.
The scale of the thing
First, the raw picture. Organised opposition to data centre development has grown faster than almost anyone in the industry expected. Trackers such as DataCenterOpposition.com counted roughly 76 local opposition groups in the United States at the end of 2025. By mid-2026 that figure had passed 400, spread across more than 40 states. Data Center Watch, an advocacy tracker, estimates that community opposition blocked or delayed tens of billions of dollars in projects over 2024 and 2025, with the pace accelerating sharply into 2026. Project cancellations quadrupled year on year, and most of them clustered in the second half of 2025.
It is not confined to America. A Chilean court suspended a Google facility after residents worked out the annual water draw. Ireland, already straining under the share of national electricity consumed by data centres, has faced years of contested connection policy. And in Cape Town, the Housing Assembly and the UK non-profit Foxglove, advised by the Legal Resources Centre, formally opposed Equinix's proposed King Air Industria development on the grounds that the land use application disclosed almost nothing about water use, electricity demand, diesel backup generation, air pollution or noise. The tribunal gave initial approval, but one member conceded that the municipal definition of a data centre was wholly inadequate, which is not the kind of remark that usually accompanies a smooth approval process.
That Cape Town case is worth holding onto, because it demonstrates the pattern in miniature.
Four genuine parallels
Symbolic condensation. GMOs were never really only about GMOs. They became the vessel for a much broader set of anxieties: about corporate consolidation of the seed supply, about the loss of smallholder autonomy, about eating food no one could quite explain. The technical question of whether a Bt maize cob was safe to eat was almost beside the point, because the argument was never fully about the cob.
Data centres occupy exactly this position now. They are the only part of artificial intelligence that most people can actually see, touch, hear humming at night and point at from the road. Everything the public feels about AI, including fears about jobs, about concentrated corporate power, about a technology arriving whether anyone asked for it or not, now has a concrete address in an industrial park. A Vox headline captured it with unkind precision: Americans do not know how to fight AI, so they are fighting data centres. Brookings has made much the same observation, describing the land use fight as the crystallisation of a broader public concern.
Transnational, networked, low budget, effective. The anti-GMO movement was one of the first genuinely networked environmental campaigns, in which a small number of well-briefed organisations supplied legal and technical capacity to a very large number of local groups. The data centre opposition looks almost identical in structure. Foxglove, a London non-profit, appearing alongside a Western Cape housing movement in a Cape Town land use tribunal, is precisely the model that Greenpeace and Friends of the Earth pioneered around GM crops. The asymmetry is stark and familiar: modest organisations with a good lawyer and a spreadsheet, up against firms deploying hundreds of billions in capital expenditure.
The demand is for disclosure before it is for prohibition. This is the sharpest parallel and the most underappreciated one. The anti-GMO movement's most durable political victory was not a ban. It was mandatory labelling. Campaigners converted a fight about safety, which they were losing on the evidence, into a fight about the right to know, which they won almost everywhere.
Look at what the Cape Town objectors actually asked for. Not a prohibition on data centres, but the water figures, the electricity demand, the generator specifications, the noise modelling. Foxglove's stated objection was that there was simply not enough information for a decision on a project of that scale. That is a labelling argument in industrial clothing, and it is strategically potent for the same reason: it is very difficult for a developer to argue publicly against telling people things.
Regulatory venue. GMO politics was fought and largely decided at the level of national regulators and supranational bodies, but its effective battleground turned out to be local: municipalities declaring themselves GM free, retailers making sourcing commitments, provinces setting their own rules. Data centre politics has skipped the national stage almost entirely. There is no equivalent of a national approval authority. Everything runs through zoning boards, planning commissions, land use tribunals and municipal electricity tariffs. That decentralisation cuts both ways, which we will come to.
Three places the analogy fails
The evidence base is not comparable. This is the crucial disanalogy, and any balanced account has to lead with it. The scientific consensus on approved GM crops, sustained across the WHO, the European Commission's own research programmes and national academies, was that they posed no greater risk to human health than conventional equivalents. That consensus allowed industry to frame opponents as anti-science, and the framing largely stuck in North America.
The data centre debate offers no such refuge. The impacts under dispute are not speculative long-term health effects. They are metered. A facility either draws 174MW or it does not. The Legal Resources Centre's submission in Cape Town used a figure of roughly 25.5 million litres of water per megawatt per year for traditionally cooled facilities, which puts the King Air Industria proposal in the region of 4.4 billion litres annually. Those numbers are contestable in their assumptions, and the industry correctly points out that many modern facilities use closed loop or air cooled systems that change the arithmetic dramatically. But the argument is about which number is right, not about whether harm exists at all. In a city that watched its taps come close to running dry during Day Zero, that is not an abstraction.
The same applies to electricity prices. Brookings has documented substantial increases in American residential electricity costs since 2019, and the question of whether ordinary ratepayers should be cross-subsidising the load growth of trillion dollar companies is a legitimate distributional question, not a scare campaign. For a South African reading this after a decade of load shedding, it is barely a question at all.
The benefit distribution is genuinely worse. GM crops delivered clear, immediate, measurable value to the farmers who planted them, which is why adoption ran ahead of public sentiment in most producing countries. Data centres deliver enormous value to their operators and to distant users, while delivering to the host community a modest rates base, a small number of permanent jobs and a set of very local costs. Even sympathetic analysts concede this. The industry's own answer, community benefit agreements, is essentially an admission that the default distribution does not work.
The politics are not the same shape. Anti-GMO sentiment mapped reasonably neatly onto a green, left leaning, urban constituency in most countries. Data centre opposition does not. It unites conservation groups with property owners worried about values, rural conservatives suspicious of outside capital, and consumer advocates focused on tariffs. In the United States it has already produced hard electoral outcomes, most notably in Warrenton, Virginia, where residents replaced an entire town council that had supported an Amazon campus. Coalitions that broad are rare and dangerous to underestimate.
What the GMO history actually predicts
If the parallel holds even loosely, three things follow.
The first is that outright national bans are unlikely, and are also not the real risk to the industry. The GMO experience suggests the binding constraint is cumulative friction: longer approval timelines, higher disclosure burdens, mandatory impact studies, water and power ringfencing conditions, and the slow hardening of what counts as an acceptable project. Death by a thousand conditions of approval is the realistic scenario, not prohibition.
The second is that geography fragments. GM crops became normal in the Americas and largely unavailable in Europe, and the split has persisted for a quarter of a century. Expect the same divergence in data centre siting, with capital flowing towards jurisdictions offering surplus power, cool climates, permissive planning and political enthusiasm, and away from water stressed or politically activated regions. South Africa sits awkwardly on this line. National government has pledged tax incentives and policy reform to attract digital infrastructure investment, while municipal tribunals in Cape Town are simultaneously discovering that their own zoning definitions cannot describe what is being built.
The third, and the one the industry should take most seriously, is that early information asymmetry is what poisons these debates permanently. The agricultural biotechnology sector's central strategic error was treating public concern as a communications problem rather than a legitimacy problem, and resisting disclosure until disclosure was imposed on it. By then the industry had no credibility left to spend. Data centre developers who arrive at a hearing without water figures are making the identical mistake, and the Cape Town tribunal's remark about inadequate definitions is a warning shot rather than a verdict.
A balanced conclusion
The comparison is useful but should not be used as a dismissal. To say that data centre opposition resembles the anti-GMO movement is sometimes intended as an insult, implying irrational technophobia dressed up in scientific language. That reading does not survive contact with the evidence. The GMO campaigners were fighting about a risk that mostly was not there. The data centre objectors are fighting about resource allocation that demonstrably is.
Where the parallel genuinely holds is in form rather than substance: the symbolic overloading of a physical object with a much larger public argument, the networked and transnational organising, the strategic pivot from prohibition to disclosure, and the near certainty that the industry will concede the disclosure fight several years later than it should have, and get less credit for it than it would have.
Both sides have something to lose by getting this wrong. A country that blocks all digital infrastructure forfeits a serious economic opportunity and hosts its own data somewhere else, at someone else's price. A country that approves it without knowing the water and power numbers is making a large bet with a scarce public resource and no way to audit the wager. The gap between those two failures is exactly where the disclosure argument sits, and it is why the objectors, in Cape Town as in Virginia, keep asking the same unglamorous question. Not whether the thing should exist, but how much it will actually use.
