Every previous technology disruption left time for labor markets to adjust. AI does not. It affects all cognitive work simultaneously, and it improves faster than institutions can adapt. The standard policy responses, defining AI in statute, mandating human-to-AI ratios, offering retraining programs, fail for a structural reason: they attempt to regulate a concept that cannot be cleanly bounded. This paper proposes a fundamentally different lever. Inspired by the Federal Reserve’s use of the federal funds rate to steer the entire economy without naming individual sectors, we identify data-center electricity consumption as the measurable, technology-agnostic proxy through which governments can modulate AI deployment speed. Power use scales with model size and inference intensity, is already metered at the facility level, and requires no legal definition of “AI.” It is the interest rate of the cognitive economy.
Implementation combines progressive taxation on compute-intensive electricity, tradable energy quotas, and employment-linked rebates, three instruments that shift the economic incentive from wholesale labor substitution toward human-AI augmentation. The framework is internationally extensible (the ECB and EU AI Act provide natural adoption pathways) and designed to evolve through the same decades of institutional learning that shaped the Federal Reserve from 1913 to the present. This paper examines why traditional approaches fail, how the electricity lever creates enforceable macroeconomic steering without micromanaging use cases, and what conditions would require the framework to adapt. The goal is not to slow AI. It is to ensure that the gains from intelligence abundance are distributed broadly enough to sustain the social contract. The alternative is technological feudalism.
