GitHub Copilot's pricing switch is the cloud compute story running again

Microsoft just moved GitHub Copilot from seat-based to pay-per-token pricing. A lot of the coverage is treating this as a monetization tweak. It's not. It's the same structural shift that happened in cloud compute, just running ten years later in AI software.

Here's the thing. Seat pricing is a hiding mechanism. When utilization is low, the vendor captures the full subscription whether users show up or not. When utilization scales, the vendor's margins compress — because delivering value at volume costs real money. Pay-per-token flips this. Usage is now legible. You can see what people are actually doing, and the vendor gets paid proportionally to the compute actually consumed. That's the honest model. It's also the model that wins long-term, because enterprise buyers eventually run the utilization math and seat pricing stops looking like a deal.

Microsoft is also staring at $190 billion in 2026 capex — roughly $32 billion last quarter alone just to bring more Azure compute online, with another $158 billion planned before year-end. Amy Hood's ROI math: $97 billion in infrastructure spend over four quarters to generate $37 billion in ARR. When you're running numbers like that, you need pricing that scales with actual usage. Seat pricing leaves too much signal on the floor. The rest of the industry will follow. They always do once the math surfaces.