Hugging Face has struck a multi-million dollar strategic partnership with Arcee AI, making the company the first major US AI lab to move all of its public and private models, datasets, and agent traces off Amazon's AWS S3 storage and onto Hugging Face.
The private hosting part of the project is powered by Hugging Face's private storage product, Buckets. Buckets is a per-TB cloud storage layer that waives egress fees, includes CDN acceleration, and is built for fast reads and writes of AI weight files and datasets. Because the storage layer is cloud-agnostic, Arcee AI becomes compute agnostic — it can freely schedule training clusters across different cloud providers without paying any data transfer fees or getting locked into a single vendor.
Arcee AI is one of the most active US labs on the Hugging Face platform. It has released 204 models and 63 datasets, with monthly downloads exceeding 100,000. Its well-known open-source projects include the Trinity series (Trinity-Mini, Trinity-Nano, and Trinity-Large-Thinking), the AFM-4.5B base model (over 160,000 downloads), Llama-3.1-SuperNova-Lite (over 100,000 downloads), and open-source datasets like The-Tome (containing 1.75 million fine-tuning samples) and agent-data for agent function calling.