When AI data centers meet the energy value filter

I have been working on Chapter 4 of my draft book, When Policy Falls Behind: Bitcoin, AI, and the Governance of Fast Systems. The chapter begins with a puzzle in Bitcoin mining. Imagine a mining operation powered entirely by renewable energy, located behind the meter, using electricity that would otherwise be curtailed. It helps finance generation, responds rapidly when the grid needs the power back, pays taxes, employs local workers, and complies with every applicable rule. Even then, opposition may remain.

My argument is that this residual friction cannot always be explained by electricity consumption, emissions, or grid performance. It arises partly from a deeper institutional question: what is an energy system for, and which activities count as legitimate claims on its resources? Bitcoin mining uses energy to create a thermodynamic commitment that secures a distributed monetary network. That output does not fit comfortably within evaluative frameworks built around factories producing physical goods, utilities supplying familiar services, or employers creating large numbers of local jobs.

One counterpoint in the chapter contrasts this problem with high-performance computing (HPC) and AI data centers. AI facilities also consume large quantities of electricity but their institutional logic appears more familiar. They use energy as an input into computational services sold to identifiable customers, and their proponents connect those services to productivity, scientific research, medicine, and national security. I had treated this relative legibility as one reason AI data centers might achieve institutional accommodation more easily than Bitcoin mining.

The August 4, 2026 episode of The Ezra Klein Show featuring Jasmine Sun complicates that comparison. It does not overturn the chapter’s institutional argument. It reveals that I need to draw a sharper distinction between an activity being legible and that activity being considered legitimate.

Legibility is not legitimacy

Sun’s reporting from Wisconsin and Michigan suggests that many opponents understand what AI data centers do. They know that the facilities provide computing capacity and some have used AI tools themselves. Their opposition does not rest solely on confusion about the technology or exaggerated claims about water consumption.

The more consequential objection is that many people do not see the output as sufficiently valuable to justify the local costs. They may use an AI system to draft an email or make an image while still regarding it as nonessential. The industry’s promises - scientific breakthroughs, higher productivity, medical advances, and eventual abundance - remain abstract, while new transmission lines, large industrial buildings, noise, land conversion, and possible electricity-rate effects are immediate and local.

That structure closely resembles the value-filter problem I describe for Bitcoin mining. The disagreement is presented through empirical claims about electricity, water, jobs, or emissions, but those claims do not exhaust it. Underneath them lies a volitional judgment about whether the activity should exist at the proposed scale and whether its benefits warrant a claim on shared infrastructure.

The episode offers a clear example of this distinction. Klein argues that if AI companies believe their own warnings about mass unemployment or catastrophic risk, then they do not merely have a marketing problem; they have a product problem. Better communication cannot resolve a conflict over the desirability of the future being offered. This is almost exactly the institutional point at the end of my mining chapter: once the value filter is explicit, the disagreement is not necessarily a misunderstanding that more information or advocacy can remove.

AI compute is therefore legible in a way that Bitcoin mining often is not. But legibility only allows people to evaluate the activity through an existing framework. It does not determine the result of that evaluation.

The relevant value filter also changes with scale

The podcast adds another dimension that will require more attention in my draft chapter. There is no single energy governance value filter operating uniformly across all decision arenas.

At the national level, policymakers may treat AI infrastructure as strategically important. Compute capacity can support leading firms, military and cybersecurity capabilities, research systems, and international bargaining power. Sun and Klein discuss countries using data center hosting as leverage to obtain access to frontier models, while constraints on domestic construction could push investment toward jurisdictions with fewer democratic or environmental safeguards.

At the local level, the calculation looks different. A community encounters the facility as a land-use decision, a large electricity customer, a source of construction work and tax revenue, and a relatively small permanent employer. National gains from AI leadership do not automatically translate into benefits that residents can see, access, or trust.

The same infrastructure can consequently pass through one value filter and fail another. It may appear essential to national economic strategy while appearing extractive to a host community. This is not simply a disagreement over facts. It reflects different positions, affected interests, spatial scales, and standards for judging public benefit.

Bitcoin mining encounters a related multilevel problem. A mining facility may provide valuable flexibility to a grid operator, revenue to an electricity generator, and security to a global monetary network, yet offer outputs that a provincial government or local community does not recognize as a sufficient public contribution. Operational accommodation at one level cannot settle political or epistemic disagreement at another.

This suggests I need some tweaks to my chapter’s framework. The value filter should not be treated as jurisdiction-specific; it can also be arena-specific within a jurisdiction. Grid operators, utility commissions, municipal councils, national security agencies, workers, and local residents can evaluate the same facility through different institutional commitments. Governance friction becomes especially persistent when no forum exists for reconciling those evaluations.

Process can turn latent friction into open opposition

My chapter concentrates on the substantive incompatibility between Bitcoin mining and inherited ideas about legitimate energy use. Sun’s reporting shows more clearly how governance process can activate and intensify that incompatibility.

Nondisclosure agreements have prevented some local officials from identifying proposed data-center customers, discussing project scale, or explaining expected electricity demand. Information still circulates through contractors, workers, and social media, but the officials who signed the agreements cannot respond openly. The result is not simply an information deficit. It is an institutional arrangement that makes secrecy, rumor, and asymmetric influence part of the public’s first encounter with the project.

The bargaining structure deepens the problem. A small municipality may negotiate with a developer whose legal, technical, and financial resources vastly exceed its own. Offers of tax revenue, infrastructure funding, or direct community benefits can then be interpreted less as evidence of mutual gain than as attempts to purchase acquiescence.

This shows that epistemic friction and governance form are not separate explanations. A weak or asymmetric process can convert a contestable project into evidence supporting a preexisting belief that the activity serves distant capital rather than the community. The value filter shapes how the bargain is interpreted, while the bargain reinforces the value filter.

The same feedback can operate in the opposite direction. Transparent project information, independent assessment, credible cost-allocation rules, and real local negotiating capacity cannot guarantee approval. They can, however, separate objections to the project’s measurable effects from disagreement over its purpose. That is precisely the function I assign to structured deliberation in the chapter - not manufacturing consensus but preventing empirical and normative claims from collapsing into each other.

A harder test for mining-to-HPC conversion

The podcast has direct implications for one of the chapter’s proposed empirical tests. Some Bitcoin miners are converting sites, power access, and parts of their infrastructure to HPC for AI customers. I describe this as a useful comparison because much of the energy setting remains constant while the activity’s institutional logic changes.

If a converted facility faces less governance friction after moving from mining to AI compute, that would support the argument that the original conflict concerned mining’s institutional logic rather than electricity consumption alone. The podcast suggests that this test remains valuable but the expected comparison is more complicated than laid out in the draft.

AI conversion may reduce one kind of friction while producing another. Regulators may find AI services easier to classify, customers may find the output easier to price, and policymakers may view the facility as strategically important. Local residents may nevertheless oppose the project because of its physical footprint, electricity demand, development process, effects on rates, limited permanent employment, or the wider direction of AI deployment.

The empirical question is therefore not just whether friction declines: it is whether its composition changes. Mining-to-HPC conversion may reduce classification friction at the implementation level while increasing distributive or political conflict at the local level. A good comparison would track who objects, which claims they make, what evidence changes their assessments, and whether opposition targets the facility’s measurable impacts or the social purpose of the activity.

There is also an important operational difference. Bitcoin mining can, under suitable contractual and market conditions, curtail rapidly and monetize electricity that would otherwise have little value. AI data centers are built around service expectations that usually place greater value on continuous access to power and infrastructure. Converting a site to a more institutionally recognizable use does not necessarily make its relationship with the electricity system more flexible or socially beneficial in every dimension.

Moratoria redistribute conflict

The episode also parallels the chapter’s discussion of geographic mobility. Local or state restrictions may stop a particular data center but they do not necessarily slow AI development. Developers can move to Texas, another state, or another country that offers cheaper power, faster approval, or fewer constraints.

Bitcoin mining has already demonstrated the same capacity for rapid relocation. China’s prohibition shifted substantial mining activity into other jurisdictions, including Kazakhstan, where the receiving electricity system struggled to absorb the demand shock. Mobility allowed the network to continue, but it redistributed governance costs rather than eliminating them.

For both industries, fragmentation is therefore a likely response to incompatible value filters. Some jurisdictions accommodate the activity, some prohibit it, and others compete to attract it. The technology continues across the resulting patchwork, while firms absorb relocation costs and host communities experience different combinations of investment, infrastructure stress, and political conflict.

The podcast adds a difficult democratic implication. If communities with strong procedural protections reject data centers, investment may move toward places where public opposition carries less weight. Local democratic control can then reduce democratic control over the infrastructure system as a whole. That does not make local consent an obstacle to be removed; it shows why uncoordinated siting decisions cannot carry the entire burden of governing a mobile, strategically important industry.

The deeper commonality

Bitcoin mining and AI data centers do not make equivalent claims on energy systems. They differ in load characteristics, hardware, customers, outputs, and relationships to the wider economy. Their political coalitions also differ, and the public may evaluate their benefits through very different beliefs about money, automation, security, and technological progress.

The deeper commonality is institutional. Both move at a pace that exceeds the ability of governance systems to classify their activities, produce credible information, allocate decision rights, and revise inherited standards of public value. Both can satisfy operational rules while generating conflict at political and epistemic levels. Both are geographically mobile enough that exclusion in one place may transfer pressures to another.

Yet the podcast also reveals an important asymmetry. Bitcoin mining’s principal output - thermodynamic commitment securing a global monetary network - is difficult to recognize within conventional energy governance categories. AI’s output is easier to recognize but the public may reject the industry’s valuation of that output, doubt that the benefits will be broadly shared, or oppose the future its builders describe.

That distinction sharpens rather than weakens the chapter’s central claim. Governance friction is not reducible to energy magnitude, environmental performance, or technical understanding. It emerges from the interaction between measurable effects, institutional process, distribution, and competing judgments about what energy-intensive digital systems are for.

The practical response is not to assume that better information will produce agreement. Governance needs to make the disagreement more precise. Communities should be able to distinguish grid effects from social-purpose judgments, immediate local benefits from national strategic claims, and opposition to a facility from opposition to the technology it enables. Only then can the conflict be managed through institutions designed for adaptation rather than compressed into another argument about whether the industry has communicated its benefits well enough.

---