Introducing HUGS - Scale your AI with Open Models

Source: HuggingFace Date: 2024-10-23 URL: https://huggingface.co/blog/hugs

Summary

Product launch (now deprecated): HUGS (Hugging Face Generative AI Services) was a zero-configuration optimized inference deployment service for running open-weights models on customer infrastructure, at $1/hour per container on AWS/GCP. Supported 13 models at launch (Llama 3.1 family, Mistral, Gemma 2, Qwen 2.5-7B). OpenAI-compatible API. Deprecated September 2025; users redirected to Dell Enterprise Hub and Azure AI Foundry.

Implications

HF as open-source ML hub. HUGS launching and then deprecating within ~12 months suggests HF tested the self-hosted enterprise inference market and chose to exit in favor of partnerships (Dell, Azure) rather than maintaining the operational burden of an enterprise SaaS product. This is a strategic signal: HF’s competitive advantage is the hub and tooling layer, not running enterprise production infrastructure.

Open-weights ecosystem health. The 1-hour deployment (vs. 1-week baseline from a customer quote) metric illustrates the real value of pre-optimized containers — regardless of who operates them. The redirects to Dell and Azure suggest the enterprise market for self-hosted open-weights inference is real enough that HF wanted to exit rather than compete.