For years, the energy cost of AI has been the industry's awkward footnote: real, growing, but easy to defer to the infrastructure teams and the utilities and the people tasked with keeping the lights on. That period is ending. This week brought a convergence of signals from Washington, Westminster, and the opinion pages that suggests the politics of AI power demand are shifting faster than anyone had predicted, and that the easy framing of AI as purely an economic opportunity is becoming harder to sustain.
In the United States, Energy Secretary Chris Wright told a Congressional budget hearing that public opposition to AI has become "a very real and growing" risk, particularly in rural communities. "The country as a whole is going very negative on AI, and this is a risk," Wright said, before adding: "It will be a loss to America if we stop this development." The framing is notable. Wright is broadly pro-AI and his department has a central role in the Trump administration's Genesis Mission, which relies heavily on national laboratories for AI development. That even he is acknowledging a "very real" and "growing" opposition suggests the political ground is shifting, not in Washington, but in the places where data centres are actually being built.
In the UK, Parliament's Science, Innovation and Technology Committee has opened a formal inquiry into low-energy computing, prompted specifically by AI's ballooning electricity demands. Datacenters already account for roughly 2.5 percent of UK electricity consumption, and demand is expected to quadruple by 2030. That creates a collision course with the government's net-zero commitments. The committee is looking at whether emerging approaches, including neuromorphic computing, which mimics the brain's architecture, and silicon photonics, which uses light rather than electrons to move data, could provide a way out. Neither technology is ready to deploy at scale. But the fact that MPs are probing them signals that the status quo, scaling existing chip architectures and hoping the grid can keep up, is no longer an acceptable plan.
The tension is structural, not incidental. Governments simultaneously want AI to drive economic growth and want to hit emissions targets. Data centres require vast and growing amounts of electricity. In most markets, that electricity still comes substantially from fossil fuels, or at best from natural gas as a "bridge." Utilities are investing accordingly: new gas generation, expanded renewables, and eventually new nuclear capacity are all being built or planned specifically to serve data centre demand. The investors funding this infrastructure see a multi-decade growth story. The communities hosting the data centres and their associated transmission lines see something different: industrial facilities that consume resources at scale and deliver benefits that flow elsewhere.
The Australian Financial Review, citing The Economist, reported this week that "growing resentment among American voters is turning AI into a political lightning rod," and that a laissez-faire approach is "no longer politically tenable or strategically wise" even within the Trump administration, which has otherwise been inclined to treat AI regulation as overreach. The shift is partly driven by capability, the models are now genuinely powerful enough that their potential downsides are visible to non-experts, and partly by the energy bills and grid pressure that ordinary people are starting to notice.
What this week's signals have in common is that they come from the political centre, not from the activist fringe. A Trump-appointed Energy Secretary acknowledging public opposition as a genuine risk, UK MPs pursuing cross-party inquiries into chip alternatives, mainstream financial press framing voter resentment as a strategic problem for the administration: these are not the arguments of AI critics. They are the arguments of people who want AI development to continue, but who can see that the current trajectory is generating backlash that could constrain it. The industry has tended to treat public concern about AI energy use as a perception problem to be managed. It is increasingly looking like a political problem to be solved.