Claim
Reselling AI inference at a cost-plus markup is a zero-margin business; the only sustainable pricing model is outcome-based, tied to value delivered rather than tokens consumed.
Mechanism
As model providers compete, inference costs compress toward commodity. A cost-plus markup shrinks in lock-step with the underlying cost, eventually reaching zero. Customers who bring their own API keys (BYOK) eliminate the markup entirely. Value-based pricing (per resolved ticket, per completed task, per generated report) decouples revenue from infrastructure cost. When inference gets cheaper, the value-based operator's margin expands rather than disappears.
Conditions
Holds when: the product delivers a measurable outcome the customer values above the inference cost (support resolution, code completion, document generation). Outcome value must be definable and consistently produced.
Fails when: the customer cannot define or measure outcome value, or when the product is primarily an interface layer rather than an outcome-generating system. Also fails when the customer's procurement process requires itemized cost-plus billing.
Evidence
Tunguz draws the distinction directly:
"Reselling inference at cost is a zero-margin business: a payment rail, not a software company."
"Cost-plus is a payment processor with a dashboard. Value-based is software."
He notes that BYOK arrangements destroy cost-plus viability entirely while leaving value-based pricing intact, forcing cost-plus operators to compete on platform value they have not yet built.
Signals
- Revenue grows as inference costs fall, rather than falling in proportion to token volumes.
- Customers pay per outcome (per ticket, per task, per report) rather than per session or per token.
- BYOK requests do not threaten the business model because the product charges for results, not compute.
Counter-evidence
Value-based pricing requires clear outcome measurement, which is difficult in creative or advisory use cases where value is subjective. Buyers in budget-constrained organizations may prefer cost-plus billing because it maps to procurement categories they already understand. Some categories (e.g., copilots that augment rather than replace human steps) resist clean outcome metrics, leaving cost-plus as the only practical option until measurement matures.
Cross-references
- Model substitution savings are reinvested in more tokens, not returned to the budget (Tomasz Tunguz on how model cost savings are reinvested in more tokens rather than returned to budget, June 2026)