Selling Surplus or Scarcity: Where AI Margin Actually Lives

The AI market has split into two categories determining who captures margin and who competes to zero. This paper introduces the surplus/scarcity binary diagnostic: surplus sellers — API wrappers, generic RAG pipelines, staff augmentation — face collapsing margins as base models commoditize, while scarcity sellers capture premium by controlling resources that resist replication. To map where margin collects, the paper presents the AI Value Stack (Infrastructure → Runtime → Governance → Workflow), organized by margin flow rather than technical architecture, elevating governance from a compliance function to an independent margin-bearing economic layer. A phenomenon termed the death of the dark room explains why: generative AI destroyed the organizational fog where IT failures hid for decades, replaced by a consumer benchmark effect — a $20/month ChatGPT subscription that makes every enterprise AI deployment instantly visible. This transparency creates the causal mechanism behind governance premiums: when AI outputs become visible to the board, executives purchase external governance not as compliance but as a liability shield no internal team can provide. The paper parametrizes scarcity half-life across Value Stack layers — infrastructure decays in 18–30 months, runtime in 12–18 months, governance compounds over 5+ years, workflow lock-in persists indefinitely — revealing that the most durable margin accrues to the layer most organizations underinvest in. This dynamic produces liability lock-in: switching a governance partner is not a migration but a catastrophe. Four diagnostic questions close the paper, reframing the CIO role from chief engineer to chief actuary — evaluating each stack layer not by what it computes, but by how it allocates risk.