CONCEPT
Thermodynamics of Computation
The physical law that every computation generates heat as waste; no efficiency gain eliminates this, only reduces the heat per operation while total heat scales with total computation.
Thermodynamics of computation names the application of the second law of thermodynamics to information processing: every energy transformation is imperfectly efficient, meaning some fraction of input energy degrades to heat. In modern GPUs running AI workloads, approximately 60-70% of electrical energy converts to useful computation; the rest becomes heat that must be removed or the chip throttles, damages itself, or fails. This is not a design flaw subject to engineering solution—it is physics. The Landauer limit specifies a theoretical minimum energy per bit-erasure operation; practical chips operate orders of magnitude above this limit, meaning substantial room exists for efficiency improvements. But efficiency improvements reduce energy per operation, not total energy when demand grows. The thermodynamic constraint is the floor beneath every AI interaction.
In The You On AI Field Guide
The second law of thermodynamics, formulated in the mid-nineteenth century through the work of Carnot, Clausius, and Kelvin, governs every energy conversion in the universe. No engine, biological or mechanical, converts input energy to useful