What's this about?
Hetzner, the German hosting provider beloved for aggressively cheap servers, is working on an LLM inference offering — renting you open-model intelligence by the token, at what everyone expects to be Hetzner-style prices.
What exactly is "inference"?
Inference is running a trained model to answer requests — the serving side of AI, as opposed to the training side. It's where the recurring costs live. Today inference is dominated by hyperscalers and specialist GPU clouds; a discount European provider entering signals the business is commoditizing.
What's the community saying?
HN 154 points, 87 comments
- Price-pressure hopes: Hetzner entering a market historically means prices drop; EU developers want a local, cheap, GDPR-friendly option.
- Skepticism about GPUs at Hetzner economics: inference margins depend on scarce hardware — can the cheap-metal playbook survive GPU supply reality?
- Open-weights synergy: cheap neutral inference only matters because good open models exist to run on it — another downstream effect of the week's open-weights politics.