The B2B software industry has run on annual subscription pricing for nearly two decades. Salesforce proved in the early 2000s that enterprises would pay monthly for cloud-hosted software instead of buying perpetual licenses, and the SaaS subscription model became the dominant commercial structure for business software. AI is now disrupting that structure, because the natural unit of AI consumption — tokens processed, queries answered, documents analyzed — doesn't map cleanly onto per-seat monthly subscriptions.
Usage-based pricing (UBP) is not new. AWS built its cloud infrastructure business on it, Snowflake's data warehouse runs on consumption credits, and Twilio charges per API call. What's new is that UBP is now moving into categories that were firmly subscription-based: contract analysis (Ironclad, Luminance), document automation (DocuSign AI), customer intelligence (Intercom, Zendesk), and knowledge management (Notion, Confluence).
The driver is straightforward: AI capabilities have highly variable consumption patterns across customers. A 50-person sales team at a small company might use an AI contract analysis tool five times a month. A Fortune 500 legal team might use it five thousand times. Charging both customers the same monthly per-seat rate means the small company overpays and the large company has little incentive to expand usage because more usage doesn't change their bill.
Usage-based pricing aligns billing with value delivered: if a customer gets more value, they pay more. For customers with predictable, high volumes, hybrid models — a committed base usage subscription with overage rates — provide cost predictability while preserving the economic logic of paying for what you use.
The challenge for software companies adopting UBP is the shift in revenue predictability. Subscription ARR (annual recurring revenue) is the metric that SaaS investors have optimized around for fifteen years: it is contractually committed, highly visible, and easy to model. Usage-based revenue is inherently variable — it grows when customers use more and contracts when they use less. This variability makes financial planning harder and investor reporting more complex, which is why many SaaS companies have been slow to adopt pure usage-based pricing even when it would better serve customers.
The emerging middle ground is "capacity" pricing: customers buy blocks of AI usage capacity upfront (at a discount to on-demand rates), similar to AWS Reserved Instances. This gives the software company revenue commitment and the customer cost predictability, while preserving the fundamental economics of usage-based consumption. Anthropic, OpenAI, and Cohere all offer committed use discounts that function this way, and vertical SaaS applications built on top of these providers are adopting similar structures.