AI Won’t Save the Grid, It’s Taking It Over

The Quiet Politics of Power in the EIA’s Recent Energy and AI Report

Shortly after announcing multiple power supply deals with data centers in 2024, Alliant Energy and other energy partners announced it would postpone the closure of its Wisconsin Columbia Energy Center coal plant. Image source: CC, Wikipedia.

A few days ago, the International Energy Agency (IEA) released its Energy and AI report, offering a sweeping technical account of artificial intelligence’s growing appetite for electricity. Packed with projections, charts, and mitigation scenarios, the report outlines the energy implications of a rapidly digitizing economy. But like many visions of our technological future, it speaks the language of systems and solutions while sidestepping a more basic question: infrastructure for whom?

A glance at the report’s peer review panel is both quite logical from a narrowly technical point of view but makes its orientation clear. Its experts hail from the major corporate players of the AI, technology, and energy sectors—IBM, Google, Cisco, Nvidia, Meta, Chevron, and more—with few critical voices in sight. Still, the report offers a useful snapshot of how energy systems are being reorganized around the demands of AI. Here are a few key takeaways worth unpacking.

---

Yes, AI consumes energy—and increasingly so. According to the report, some data centers already draw as much electricity as 100,000 households. The largest now in planning could use 20 times that. By 2030, global data center energy demand is set to more than double, exceeding the total electricity consumption of Japan. But this isn't merely a matter of load balancing or energy efficiency. At stake is a structural transformation—where public energy systems are being reorganized around the operational imperatives of private computational empires.

Yet absent from the report is any sustained reflection on the monopolistic dynamics driving this transformation — namely, how a handful of firms are not only consuming energy at unprecedented scales but also reshaping the governance of energy systems themselves. While the IEA report notes that companies like Microsoft, Amazon, and Google are “planning as much as USD 300 billion in AI-related capital expenditure in 2025” — exceeding total U.S. power sector investment — it portrays them as constructive market participants rather than dominant infrastructural actors reorganizing territory and resource flows around their operational needs. This framing overlooks a critical reality: the privatization of infrastructure transforms energy systems from public goods into corporate assets, reconfiguring them to generate profit while severing essential services from democratic oversight and insulating them from public accountability.¹

Indeed, platform giants aren’t just large energy consumers; they’re becoming de facto energy planners. Through long-term power purchase agreements, private grid investments, and proprietary energy optimization platforms, they’re hardwiring their operational needs into public infrastructure while shielding themselves from regulatory control. What we’re witnessing is not simply innovation, but infrastructural capture.

---

This capture has geography. Data centers cluster in places like Northern Virginia and Arizona—not by chance, but because of tax breaks, cheap water, and pliant regulation. The IEA report notes that by 2035, U.S. data centers may draw as much peak power as the entire industrial sector. That figure isn’t just impressive—it’s diagnostic. When industrial capacity is redirected toward serving cloud infrastructure, we’re no longer just talking about “the cloud.” We’re talking about an energy regime organized around platform capitalism.

The state, for its part, is far from a neutral referee. Whether under Biden’s push to streamline clean energy access for AI data centers, or Trump’s order to reinvigorate “America’s Beautiful Clean Coal Industry” for AI data centers, the state is an enabler. Echoing Timothy Mitchell’s argument in Carbon Democracy, these infrastructural choices shape not just economies, but the very distribution of political power—privileging those who can build at scale and lobby accordingly.²

The IEA predicts $4.2 trillion in global data center investment by 2030, with another $480 billion in power sector upgrades—half of it in the U.S. These aren’t market inevitabilities; they are political projects. And yet the report treats them as natural outcomes of technological progress, not as contested developments with winners, losers, and locked-in consequences. Underlying this is not a rational outlook on technology but a belief that AI advancement will simply solve the problems it is helping to exacerbate.

---

The vision offered to emerging economies is equally familiar. The report suggests that with the right governance, countries in the Global South could “leapfrog” into AI-enabled energy optimization. But these leapfrogs often land in the same old potholes: extractive dependencies, outsourced emissions, and lingering infrastructural inequities.³ Despite representing half of all global internet users, emerging economies host less than 10% of global data center capacity. This isn’t a bug—it’s the result of a deeply uneven computational geography.

And even when these countries are included, it’s often as sources of critical minerals, not partners in digital infrastructure. By 2030, data center demand for gallium alone may reach 11% of current global supply—nearly all of it refined in China—which is currently the target of a U.S. trade war. So while AI is sold as a path to decarbonization and development, it’s tethered to the same geopolitical logics and material asymmetries that have long shaped the global energy economy.

---

To its credit, the IEA report flags potential delays due to permitting, supply chain stress, and grid bottlenecks. But it does so with faith in technocratic solutions. Emissions from AI infrastructure are projected to rise—data centers alone may produce 3% of global power sector emissions by 2030—yet the report remains confident that digital optimization will save more than it burns. This optimism sidesteps a thornier truth: energy governance today remains dominated by fossil interests, fractured public institutions, and corporate monopolies. That’s a shaky foundation for an AI-powered climate fix.

To be clear, the IEA report does not necessarily reflect the personal views or politics of its individual contributors or reviewers. But it’s worth reflecting on how issues of economic concentration are framed—or sidelined—in technical assessments of energy and AI. In a scholarly article published in Nature Climate Change and co-authored by one of the IEA report’s reviewers, the problem is acknowledged, but only briefly:

"ML expertise today is often concentrated among a limited set of actors, raising potential challenges with respect to the governance and implementation of ML in the context of climate change," including the risk of "shifting power from public to large private entities by virtue of who controls relevant data or intellectual capital."

The point isn’t that this concern is absent — it’s that it rarely shapes the analysis in proportion to the scale of the problem.

---

In the end, Energy and AI is less a roadmap than a snapshot of how infrastructural priorities are being set—with minimal public debate and maximal private gain. The report sees infrastructure as a system to be optimized. But infrastructure is never just pipes and wires—it is the spatial and political architecture of society. It determines who gets power, in every sense of the word.

The IEA is right that AI and energy are converging. But the real question isn’t how to feed AI’s energy hunger. As climate crises deepen and resource constraints tighten, the question is whether we can reclaim infrastructure—energy, data, and territory—not as a substrate for private empire, but as a site of collective negotiation over how we live and what futures we build. As many who study infrastructure remind us, infrastructure is never just about pipes, wires, or servers—it is about promises made and unmade, about visions of progress that materialize unevenly and often fail.⁴ The challenge ahead is not to meet AI’s demands more efficiently, but to ask: whose promise will our infrastructures serve, and at what cost?

References

1. Christophers, Brett. Rentier Capitalism: Who Owns the Economy, and Who Pays for It? Verso Books, 2020. Also see Dawson, Ashley. People’s Power. New York, NY: OR Books, 2022.

2. Mitchell, Timothy. Carbon Democracy: Political Power in the Age of Oil. Verso Books, 2013.

3. See Larkin, Brian. “The Politics and Poetics of Infrastructure.” Annual Review of Anthropology 42, no. 1 (October 21, 2013): 327–43. Ferguson, James. Global Shadows. Durham, NC: Duke University Press, 2014.

4. Anand, Nikhil, Akhil Gupta, and Hannah Appel, eds. The Promise of Infrastructure. Duke University Press, 2018.