Hugging Face AI Storage Buckets

Hugging Face Storage Buckets is object storage for storing models, datasets, checkpoints, and other AI artifacts. Files can be synced instantly without Git commits, and every bucket includes a CDN.

The core feature is Xet deduplication. It splits files into byte-level chunks using content-defined chunking and removes duplicate data across the entire bucket, so retraining that changes only part of the model weights only re-uploads the changed data. In an example environment, analyzing 2.4TB and 12,847 files deduplicates 62% and uploads only 912GB.

Datasets can consolidate crawled data, annotations, synthetic data, and more into a single bucket. It handles large volumes of files without commit queues or file count limits, and the same data can be used across any compute environment you want, including AWS, GCP, Nebius, CoreWeave, Lambda, or your own cluster.

Pricing is usage-based, at roughly $8-18 per TB per month for storage. Public repositories start at a default of $12/TB per month, and private repositories start at $18/TB per month, with discounts of up to 33% applied at 50TB, 200TB, and 500TB+ usage tiers. Egress and CDN usage are not charged separately, and the deduplication billing benefit for original and processed data requires an Enterprise or Enterprise Plus plan.