Cheaper tokens only help infra if volume moves.
Turn a market narrative into unit economics. Every value is your assumption; no ticker, provider margin, or future outcome is treated as live fact.
Scenario arithmetic only. Not investment advice, a valuation, a forecast, or verified economics for NBIS, NVDA, ORCL, NOW, Meta, OpenAI, Anthropic, Apple, Google, Grok, or any other company.
The thesis has a denominator.
Infrastructure dollars rise when future volume times infrastructure capture exceeds the baseline. A lower model price alone does not change that multiplication.
USD 200
stays USD 200
at 1.0x
The no-growth counterfactual exposes the hidden condition.
Price
What the customer pays per million tokens.
Capture
What infrastructure receives per million tokens in this simplified stack.
Margin
What remains after infrastructure and other model costs.