Netflix Publishes In-House LLM Serving Architecture

Netflix published its own account of how it runs large language model inference in-house, per the Netflix Technology Blog. Self-hosting LLM serving instead of paying per-token through a vendor API is a build-versus-buy call. Netflix documenting that choice in public gives other engineering teams a real case to weigh against their own infrastructure decisions.

The post reached Hacker News on August 10, the same day a run of local-first, self-hosted tools showed up in Show HN. Different scale, same instinct: run the model yourself instead of routing every request through someone else's endpoint.

What's missing is confirmation. The full post wasn't retrievable, so cost figures, request volume, and the list of models it covers aren't confirmed as a result. The claim that Netflix serves LLMs in-house rests on the post's existence and its headline, not on verified detail inside it.

Expect self-hosted LLM serving to spread among large infrastructure teams first, not indie builders, over the next year.