The billing change itself isn’t news. GitHub’s April 27 announcement gave enterprises more than five weeks of lead time before the June 1 transition from Premium Request Units to GitHub AI Credits at $0.01 per credit, per GitHub’s official vendor pricing. Two registry briefs covered the mechanics in detail before the switchover. What’s new is the two-week mark: the point where production usage data replaces budget projections, and the gap between them becomes visible.
The pattern is documented and consistent across platform billing transitions. Flat-rate subscriptions set an implicit usage floor and ceiling, teams know what they’re paying regardless of how much they use the tool. Credit-based systems expose the actual cost of heavy usage for the first time. Developer teams that ran high request volumes under the old Premium Request Unit system are, in some cases, discovering that their actual usage translated to credit consumption that exceeds what the flat-rate equivalent implied. That’s not a surprise in retrospect. It’s a surprise in practice.
The broader shift, away from flat-rate AI subscriptions toward usage-based billing, is confirmed as a documented industry trend. The Adyen acquisition of Orb for $335 million on June 11 was explicitly framed as a bet that usage-based billing infrastructure would become the enterprise AI cost management layer. OpenAI and Anthropic have made analogous pricing structure moves in their enterprise tiers, supported by prior registry coverage of Claude and ChatGPT enterprise billing changes. GitHub’s transition is the most visible current example because it affected a large installed base on a specific date.
The real story isn’t the sticker shock itself. It’s that enterprise AI governance frameworks built around headcount-based licensing, a fixed number of developer seats at a fixed monthly cost, aren’t designed for consumption-based variance. A team of 50 developers on a flat-rate AI coding tool has a known monthly cost. The same team on a credit-based system has a cost that varies with productivity, deadline pressure, and which models they’re calling. That’s a budgeting and approval process problem as much as a pricing problem. CFOs who approved AI tooling budgets in Q1 based on per-seat costs are, in some cases, seeing different numbers in the first June invoices.
What to Watch
What to watch
the first enterprise renewal cycle post-transition tells you whether teams absorb the cost difference, negotiate renegotiated terms, reduce AI tool usage, or shift to on-premise or open-weight model alternatives. The brief on the local LLM substitution effect from May 31 documented early signals of the last option. Watch whether those signals persist or whether enterprise teams find usage caps and governance controls that make credit-based billing workable within existing budgets. The Q3 enterprise renewal data, typically available in vendor earnings calls, is the first hard number that resolves this question.