Arcee Becomes the First Major American AI Lab to Replace AWS S3 with Hugging Face Private Storage, in a Multi-Million Dollar Commercial Partnership
Today we're super excited to announce the signing of a multi-million dollar strategic commercial collaboration with Arcee AI, one of the fastest-rising American AI labs. Going forward, Hugging Face will be the exclusive home for all of Arcee's models, datasets and agent traces: not just their public releases, but their private models, proprietary datasets & agent traces too. Every weight they train and every dataset they curate, open or confidential, will be stored on and distributed through the Hugging Face Hub.
That second part is worth pausing on. Plenty of companies publish their open models on Hugging Face. Arcee is going further: trusting the Hub as the production storage layer for their entire catalog, including the private artifacts their enterprise business runs on.
The private side is powered by Buckets, our private storage product, which Arcee helped shape as a launch partner from day one. Buckets gives teams simple per-TB storage with egress and CDN included, optimized for AI artifacts: fast reads and writes of weights and datasets from anywhere. Because it sits outside any single cloud, it makes Arcee fully compute agnostic: they can train on any provider, spin clusters up and down wherever capacity is cheapest, and their models and data follow them with no egress fees and no lock-in. Storage is the home. Compute is wherever you need it that day.
Why this matters
Something important is happening in AI right now: the gap between closed frontier labs and open-source builders is closing fast, and the question of who builds the leading open models matters more than ever.
The most downloaded and most remixed open models of the past year have increasingly come from outside the United States. That's great for global open science, and we celebrate it. But America's open-source AI ecosystem needs champions too. Companies that don't just consume open models, but train them, release them, and bet their business on openness.
Arcee is exactly that kind of company. While others write press releases, Arcee ships weights. Their small, specialized models are deployed inside some of the largest enterprises in the world, proving every day that you don't need a trillion parameters behind an API to deliver state-of-the-art results. You need great training, great data, and a distribution platform the whole world already builds on.
Arcee on the Hub, already
This partnership formalizes what the community has been seeing for two years. The arcee-ai organization is already one of the most active American labs on the Hub:
- 204 models and 63 datasets published as a verified organization
- Over 100,000 model downloads per month across the catalog, and millions all-time
- Flagship open releases the community knows well: the new Trinity family (Trinity-Mini, Trinity-Nano, Trinity-Large-Thinking), the AFM-4.5B foundation models (160K+ downloads for the base model alone), and earlier hits like Llama-3.1-SuperNova-Lite (100K+ downloads) and the Virtuoso series
- Open datasets that other labs train on, including The-Tome (1.75M curated instruction samples), agent-data for function calling, and the Llama-405B-Logits distillation dataset that helped train INTELLECT-1
Arcee doesn't just publish on the Hub. They build in the open, and the ecosystem builds on top of them.
What's in the partnership
- Exclusive storage and distribution, public AND private. All Arcee models and datasets will live on the Hugging Face Hub: open releases distributed to the world, and private models and proprietary datasets stored securely in private repositories, with egress and CDN included at full speed.
- Arcee as a flagship organization on the Hub. Expect deeper integrations, featured releases, and co-developed drops throughout the year.
- A shared commitment to open American AI. More open weights, more open datasets, more reproducible results, released where 15 million AI builders can use them on day one.
For Arcee's enterprise customers, this means one trusted, battle-tested supply chain for models: the same platform that serves billions of downloads a month, now the single source of truth for every Arcee artifact.
"Arcee is one of the best examples of what American open-source AI can be: small teams shipping world-class open models that enterprises actually run in production. Making Hugging Face the exclusive home for their models and datasets means the whole community benefits from everything they build next."
Clément Delangue, co-founder & CEO, Hugging Face
"Hugging Face is where AI lives. Making it the exclusive home for everything we build, from our public releases to our private models and proprietary datasets, was an easy call: it's the best infrastructure for AI artifacts, and the fastest way to get our work into the hands of every developer and every enterprise on the planet."
Mark McQuade, co-founder & CEO, Arcee AI
What's next
The first co-released drops are already in the works. Follow arcee-ai on the Hub to be first in line!
