GitHub Copilot's shift to a consumption-based billing model is, at its core, a public confirmation of something the industry already knew privately: the "unlimited" AI subscription model was never financially sustainable. The core problem was clear — a multi-hour autonomous coding session cost GitHub the same as a simple chat question. The company was quietly absorbing the difference, until it no longer could.
What strikes me as most significant here isn't the change itself, but the signal it sends across the entire industry. Anthropic and Google had already taken steps to limit usage of their services before GitHub's decision, and OpenAI is considering ending unlimited plans — all of which suggests we're entering a phase of economic maturity in AI, where the excitement of the early years gives way to business models that actually need to make sense in the real world.
For users, the most legitimate concern is unpredictability. Usage-based billing is non-deterministic: you can never know exactly how many tokens a model will consume to respond to a given input. That makes budgeting genuinely difficult, especially in organizational settings. The promise of a "bill preview" experience in May is a step in the right direction, but it doesn't eliminate the structural problem.
The buffet is over. The question now is whether the new à la carte menu will keep developers at the table.
GitHub Copilot's shift to a consumption-based billing model is, at its core, a public confirmation of something the industry already knew privately: the "unlimited" AI subscription model was never financially sustainable. The core problem was clear — a multi-hour autonomous coding session cost GitHub the same as a simple chat question. The company was quietly absorbing the difference, until it no longer could. What strikes me as most significant here isn't the change itself, but the signal it sends across the entire industry. Anthropic and Google had already taken steps to limit usage of their services before GitHub's decision, and OpenAI is considering ending unlimited plans — all of which suggests we're entering a phase of economic maturity in AI, where the excitement of the early years gives way to business models that actually need to make sense in the real world. For users, the most legitimate concern is unpredictability. Usage-based billing is non-deterministic: you can never know exactly how many tokens a model will consume to respond to a given input. That makes budgeting genuinely difficult, especially in organizational settings. The promise of a "bill preview" experience in May is a step in the right direction, but it doesn't eliminate the structural problem. The buffet is over. The question now is whether the new à la carte menu will keep developers at the table.