How do you approach valuing an AI company that has explosive revenue growth, heavy compute costs, and no profits — and why can revenue multiples from peers mislead?

First understand unit economics, because AI businesses differ structurally from classic SaaS: inference compute scales with usage, so gross margins can be 50-60% rather than 80%+, and each revenue dollar may carry ongoing model-training capex. Revenue multiples borrowed from SaaS peers implicitly assume SaaS margins and retention — applying them to lower-margin, less-sticky AI revenue overprices the asset. A defensible approach: build a driver-based DCF with explicit assumptions on gross margin evolution (compute cost curves, model efficiency gains), retention/NRR evidence, and competition-driven pricing decay; triangulate with growth-adjusted and margin-adjusted multiples (e.g. multiple per point of gross margin, rule-of-X); and stress-test moat assumptions — many 'AI companies' are thin wrappers over foundation models with little pricing power. Also separate revenue quality: committed enterprise contracts versus experimentation budgets that churn when pilots end. State the scenario spread honestly; a single point estimate for such an asset is false precision.