# AI Foundation Models Are Energy **Thesis:** AI foundation models function like energy in the economy — a fundamental input that everything else runs on. The companies building the best foundation models are the energy producers of the AI era. **Date:** 2026-03-12 **Author:** Junwon Park --- ## The Power Plant Analogy Traditional energy: oil and electricity power heat and motion. Foundation models are power plants for cognition — input electricity, output tokens. Tokens are analogous to watts: a measure of cognitive capacity or work that can be done. A million tokens from a newer model produces better work than a million tokens from an older one, the same way a kilowatt-hour from a modern combined-cycle gas turbine represents more efficient conversion than one from a 1960s coal plant. But within a given generation, tokens are a reasonable unit of cognitive output. ## Structural Parallels **Massive capital expenditure upfront.** Power plants cost billions before generating a single kilowatt-hour. Foundation model training runs cost hundreds of millions to billions before producing a single useful token. The economic moat is the willingness and ability to spend at that scale. **Economies of scale.** Once built, the marginal cost of each additional unit of output drops. A power plant's economics improve with utilization. Foundation models improve with inference volume — the training cost is fixed, and every additional API call amortizes it further. **Output is largely fungible at a given quality tier.** Electricity from one plant is interchangeable with electricity from another (same voltage, same frequency). Tokens from competing frontier models are increasingly interchangeable for most tasks — the differentiation is in cost per token, latency, and reliability, not in fundamental capability gaps. **Vintage matters.** Newer plants are more efficient per unit of input. Newer models produce more useful work per token. The competitive pressure is to build the next generation before your current one becomes the equivalent of a coal plant in a natural gas world.