If you’ve been on Twitter lately, you’ve seen Anthropic enforcing people to use Claude Code for their subscriptions, blocking tools like opencode from using it.
Most videos and breakdowns I see say they’re doing it to break even on heavy users who abuse the max plan, but honestly it’s a half truth when we truly zoom out.
The entire industry is currently extremely subsidized by venture capital, we aren’t even close to paying the real cost.
The actual price of inference and model training is ASTRONOMICALLY higher because of the insane burn rate on infra, chips, and how non-sticky new models are, because there’s another one released every other week.
I think the bubble will “pop” not because AI doesn’t work, but when the industry gets impatient and the prices start to actually rise to meet reality - and people lose interest in paying that much.
It could be 2026, it could be 2027. But honestly it seems sooner than later.