Cheap cloud AI was never going to last. The off-ramp wasn't built for everyone. (opens in new tab)

On June 1, GitHub Copilot moves to usage-based billing. On June 15, Anthropic caps Claude Code. The flat-fee era is ending in a two-week window. Heavy users have been consuming roughly $8 of compute per $1 of subscription revenue; the labs aren't covering that anymore. The off-ramp – self-hosting open-source models – exists, but charges a skill toll that excludes the user base the labs onboarded. Who gets left where, and was the door ever meant to stay open?