Hetzner is working on LLM Inference (sliplane.io)

Hetzner is currently experimenting with Large Language Model (LLM) inference through its newly unveiled API, which is compatible with OpenAI’s infrastructure. This early-stage project allows users to test the Qwen/Qwen3.6-35B-A3B-FP8 model, a 35-billion-parameter Mixture-of-Experts model capable of processing text and images. However, Hetzner emphasizes that this is strictly a trial phase—there's no billing, service level agreement (SLA), or production reliability, and it serves primarily to gather insights on user interest and system scalability. The significance of Hetzner's initiative lies in its potential to disrupt the LLM inference market, which is becoming increasingly commoditized. With its efficient infrastructure and hardware acquisition capabilities, Hetzner could position itself as a viable low-cost alternative for LLM services if it expands its offerings beyond smaller models. Currently equipped with robust GPUs suitable for medium-sized models, Hetzner's next steps will be critical; if it develops larger GPU clusters and more comprehensive model selections, it could emerge as a serious contender in the inference space, particularly in Europe where it boasts an established reputation for competitive pricing.