Intelligence has always been scarce. Artificial intelligence is changing that. For the first time, cognitive labour — analysis, synthesis, reasoning, composition — can be generated at industrial scale, at near-zero marginal cost. This has encouraged a widespread assumption: more intelligence will naturally produce better outcomes. The evidence says otherwise. This paper argues that intelligence generation and intelligence verification are distinct functions, and that verification is proving far more resistant to automation than generation. The resulting gap — the Verification Deficit — is the growing disparity between a system's capacity to produce outputs and anyone's capacity to determine whether those outputs are true. The deficit is not new. Medicine practised bloodletting for two thousand years. Financial markets inflate and collapse on a recurring schedule. Engineering disasters follow from institutional failures to verify what was already known. In every case, the constraint was not intelligence. It was verification. Artificial intelligence did not invent this failure mode. It industrialised it. The paper introduces Cost Per Verified Outcome (CPVO) as an economic framework for measuring what intelligence actually costs when you include the price of determining whether it worked. It argues that verification is emerging as the foundational infrastructure layer of the intelligence age — as essential as communications networks were to the information age and energy systems were to industrialisation. ---