Many (most?) people are not at that kind of scale, and LD is only at scale as the sum of 1000s of small sites + a few big ones.

After using LD for a year I found it expensive, slow, broken. They were cool and gave me a large credit.

Now I use a self-hosted "microservice" which works at our scale (100s of users).

Feature Flags are only checked on sign in here, so the load on our service is very low.

But is in go, which is "fast" and web-scale, right?

Wow, $79-300 a month is expensive for a feature like that. Not sure who they're targeting at a price point like that - many who can cough up a budget like that for a small point solution are going to have the capacity to build their own.

split.io is several times that.

IDK about this product, but split.io has more features and better UI than you could get from a handful of engineers working part-time.

If one is "at scale", they will have enough in-house resources to implement this by themselves.

That doesn't mean they will though. Companies frequently make the decision to buy services that they could build instead. I'm not saying every one should be paying for feature flagging, just that it could make sense for some companies.