Hugging Face launches Train on DGX Cloud for Enterprise Hub — no-code GPU training on NVIDIA DGX Cloud

AI Impact Summary

The deprecated DGX Cloud training path has been replaced with Train on DGX Cloud, a Hugging Face Hub service for Enterprise Hub organizations. This enables no-code fine-tuning of models using NVIDIA DGX Cloud GPUs (H100, with L40S coming soon) via AutoTrain and Spaces, outputting private model repos in the user namespace. Enterprises must manage entitlements, per-minute GPU costs, and dataset formatting (CSV/JSON) while leveraging architectures like Llama, Falcon, Mistral, Mixtral, T5, Gemma, Stable Diffusion, and Stable Diffusion XL to accelerate open-model training.