Hugging Face has announced a partnership with SkyPilot, an open-source tool for running AI workloads across multiple clouds. The integration allows developers to use Hugging Face as a primary storage layer while executing compute on any major cloud provider, such as AWS, GCP, or Azure. This means data can be stored once on Hugging Face and accessed from wherever training or inference runs, without needing to copy it between clouds.
The key benefit is eliminating egress fees, which typically apply when data leaves a cloud provider's network. By keeping data on Hugging Face and using SkyPilot to orchestrate compute, teams can avoid those charges. The setup also simplifies workflow: SkyPilot handles the job scheduling and cloud selection, while Hugging Face handles versioning and dataset management.
The source describes the feature as "zero-egress" and highlights that it works with existing Hugging Face datasets and models. No mention is made of performance trade-offs or limitations, so the article reflects only the positive claims from the announcement. For teams already using Hugging Face and multi-cloud strategies, this could reduce cost and operational overhead.