The AI subsidy era is ending

For the past couple of years, building with AI felt almost free. Anthropic and OpenAI were burning money to get developers hooked, and it worked. Nobody really talked about costs.

That's changing now. Business accounts are moving to real API pricing, and the jump is significant enough that cost is suddenly a first-class concern.

I'm seeing teams explore things they largely ignored before — complexity routers that classify requests and route them to cheaper models, open-weight models for tasks that don't need frontier quality, hybrid setups that only hit the expensive APIs when necessary.

It's a healthy forcing function. A lot of AI features got built without anyone asking whether the unit economics made sense. Now they have to.

That said, I don't think frontier model prices are going to collapse anytime soon. The compute costs are real, and the labs have their own pressures. So this isn't a problem that just resolves itself in six months.

The teams that figure out smart model routing and tiering now will have a real advantage. Not just on margins — on the discipline of knowing which tasks actually need the best model and which ones don't.