We propose the Complex Market Hypothesis: substantial return predictability arises from many individually weak signals. We model a computational arms race in which investors choose model scale and learn from limited data. Random matrix theory yields an essentially unique linear equilibrium and characterizes the distribution of model sizes. Prices transmit information from larger models to smaller ones, allowing both to coexist. Cheaper computation can raise aggregate spending and widen performance gaps across funds. Even when every investor observes all signals determining payoffs, uncertainty about their predictive coefficients sustains a sizable equity premium.