I get the argument for owning your data and having everything on your infrastructure. But that approach can never outscale and outpace what an API provider is doing. They have entire teams, hundreds of engineers, dedicated to making the models better, faster, cheaper. One person or a small team trying to keep up with that. Can't. The gap gets bigger over time.

The costs, the optimization, the tinkering. It's the fun part for most technical people. It feels like work. It feels like you're building something. But what you're actually doing is maintaining something. And nothing wrong with that. However, there's usually a persistent need to tune the model output, to keep things running, to patch things when the next update breaks something.

The energy put into all of that could have gone somewhere else. Into building on top of APIs. Someone else could run it for you at a fraction of the total cost when you actually account for everything.

At the end of the day, it's preference. 

Open source people like the infrastructure. How things work, control over every piece, tweaking and optimizing. The model is the project. API people see the model as a tool. Build something on top and move on. The product is the project.

Neither is "saving money". It's a stupid argument. Time is money. And people forget to count the 40 hours they spent keeping it all together.

It's not a critique, just an observation. I love falling into making my own infrastructure, but it can never be as efficient as an API.