AI Pricing

Pricing AI software is hard because the value metric is moving faster than the pricing models are.

Three distinct problems are colliding. The technology produces value through different mechanisms than traditional software, and cost-to-serve scales with usage in ways subscription pricing can't absorb. Buyer willingness to pay is bound to the buyer's own ability to extract value, which depends on workflow integration, change management, and accuracy thresholds. The category is repricing under load.

In the thirty days before this hub launched, GitHub, Atlassian, and HubSpot all repriced their AI products. Three different metric bets, one shared underlying problem. Vendors are watching each other and shifting bets every few weeks because no one has settled on what the unit of value actually is.

"Credit-based," "outcome-based," and "consumption-based" pricing aren't competing pricing models. They're three different bets on what the value metric should be — with the pricing-model debate masking a value-metric debate one layer upstream.

The visible debate is the pricing model; the actual disagreement lives upstream, in the licensing model (where the value metric lives). SPP analyzes AI pricing at the metric layer because the pricing model is downstream of the metric — and the packaging model (how licensed units bundle into editions or tiers) isn't where the AI debate is yet. Get the metric wrong and no pricing-model choice saves it.

Start with the overview below — it frames the structural problem at the metric layer. The articles that follow cover each specific bet, the failure modes already visible across GitHub Copilot, Atlassian Rovo, HubSpot Breeze, and recent GenAI repricings, and where decomposing AI products into model, agent, and workflow layers resolves apparent contradictions. This hub doesn't cover non-AI pricing models (see SaaS Pricing) or value-based-pricing methodology in general (see Value-Based Pricing).