The Good Citizen Paradox

Why Suppressing the Data Center Price Signal Prolongs the Very Scarcity It Tries to Manage

A Different Debate

At CERAWeek 2026 in Houston, two of the sharpest minds in the data center energy space squared off on stage. I wasn’t there, but by all accounts, it got heated.

Amanda Peterson Corio, Google’s global head of data center energy, argued that data centers should stay connected to the grid, flex their power use during stress events, and pay for infrastructure upgrades that benefit everyone. Toby Neugebauer, CEO of Fermi America, argued the opposite: build behind-the-meter gas and nuclear generation, skip the interconnection queue, and stop burdening ratepayers with the costs of AI’s appetite.

They both made good points, but they were answering different versions of the same question: how should data centers relate to the energy system? Connect or disconnect? Gas or renewables?

Here’s a different question, and I think it’s the one that unlocks the others: what can this demand class actually pay for electricity?

The Missing Variable

The Compute Heat Rate answers this directly. Derived from GPU economics, CHR¹ quantifies the maximum electricity price each tier of AI data center workload can profitably sustain. It converts AI compute revenue into an electricity-equivalent tolerance ceiling, expressed in dollars per megawatt-hour.

The blended CHR across all major workload types is roughly $6,350/MWh. That’s about 127 times the natural gas heat rate. Individual workload tiers range from commodity inference at the low end (around $250/MWh) to frontier model training and premium inference at the high end ($49,000 to $74,000+/MWh). The full methodology is published on SSRN.

Two industries are talking about the same electrons in completely different languages. CHR is the conversion factor between them.

The Good Citizen Paradox

There’s a paradox at the center of the data center energy debate, and it’s hiding in plain sight.

Data centers can sustain electricity prices orders of magnitude above what they actually pay. So they negotiate fixed-rate PPAs at $50/MWh, sign ratepayer protection pledges, build behind-the-meter generation that never touches the wholesale market, and invest in flexibility software that blunts scarcity pricing during stress events. Each of these moves is individually rational and often genuinely well-intentioned.

Collectively, they produce a perverse outcome: the grid sees incremental load growth, not transformational willingness to pay. The investment signal gets muted. The supply response comes slower, smaller, and less precisely targeted than it would if the true price signal were transmitted.

Peter Perri III captured this well in his recent piece “Houston, We Have A Solution,” describing how “the price signal is broken.” I think he’s right.

This is the Good Citizen Paradox. The more effectively data centers suppress their price signal, the longer scarcity persists, and the worse the outcome for everyone, including the ratepayers the suppression was designed to protect.

The Demand Composition Ratchet

The ratchet only turns one direction. That’s the thing to understand first, and then I’ll explain why.

When wholesale electricity prices spike, traditional industrial loads curtail. Aluminum smelters shut down potlines at $60 to $80/MWh. Steel arc furnaces pull back at $100 to $160/MWh. Chemical plants defer batch processes. These are the loads that have historically put a ceiling on price spikes by reducing consumption as prices rise.

Data centers don’t curtail. At $250 to $6,350/MWh, no market-relevant price spike triggers a rational curtailment decision. The economic lock is reinforced by a contractual lock: SLAs with 99.99%+ uptime requirements and liquidated damages clauses that make curtailment a breach, not a choice. I’ve described this elsewhere as the “dual-lock” model of demand-side inelasticity.

Now watch what happens over time. Every price spike that pushes traditional load off the system while data center load persists changes the composition of the remaining demand pool. The pool becomes more high-CHR-dominated. The curtailable load that used to provide the brake is gone. The load that remains has no price-responsive ceiling below $250/MWh at minimum.

The next spike starts from a higher floor. And the one after that starts higher still.

Even if some data center workloads flex during stress events, the ratchet continues as long as the non-flexible share grows faster than the flexible share. The economic trajectory of AI compute, with inference scaling faster than training, suggests it will.

The $6,350 Question

The objection I hear most often is some version of: “If data centers don’t actually pay $6,350/MWh, what’s the point?”

Engineers measure the tensile strength of steel not because every bridge operates at maximum load, but because you need to know the breaking point to design everything else. The measurement matters precisely because the system operates below the limit.

CHR is the tensile strength of data center demand. It doesn’t predict what data centers will choose to do. It measures what they can afford to do. And that number keeps going up, not down.

The debate at CERAWeek was about structure. Important questions, all of them. But structure without knowing the economic capacity of the demand driving it is engineering without load specs. CHR fills that gap, and the distance between what data centers can pay and what they do pay is where the most important energy policy questions are waiting to be answered.

Hans Royal is the originator of the Compute Heat Rate™ (CHR) framework. All views are his own and do not represent those of any employer or affiliated organization.

1. Royal, Hans, The Compute Heat Rate: Quantifying AI-Driven Electricity Price Tolerance
and Its Implications for Wholesale Market Repricing (February 28, 2026).
Available at SSRN: http://dx.doi.org/10.2139/ssrn.6322318