Hugging Face

Hugging Face's homepage doesn't tell you a founder origin story. What it shows instead is a list of open source libraries the team has built and maintained with the community. Its flagship PyTorch model library sits at 167,600 GitHub stars, Diffusers at 34,705, smolagents at 29,800, and PEFT at 21,780. The company's own line is "We are building the foundation of ML tooling with the community," and that's the most honest summary of who's behind this: a team that shipped the plumbing first and then built a hosting platform around the people already using it.

That platform, at huggingface.co, is where the ML world parks its work. You'll find 2M+ models, 500k+ datasets and 1M+ demo apps called Spaces. Big labs publish there alongside hobbyists. Google has 1.14k models and 70.2k followers, AI at Meta has 2.36k models, and Microsoft has 547. The trending list on any given week puts google/embeddinggemma-2 next to a community-quantized Qwen-Image fork, which tells you both what's great and what's messy about it. The site claims more than 400,000 organizations use Hugging Face.

The business model is clear once you look past the free hub. Public hosting is unlimited and free. Money comes from compute (GPU Spaces and Inference Endpoints starting at $0.60/hour), a Team & Enterprise plan starting at $20/user/month for SSO, audit logs and private dataset viewing, and Inference Providers, which routes you to 45,000+ models through one API with no service fees. Paid plans start at $9/mo on the listing. If you're building anything with open models, you'll end up here eventually. The real question is how much of your stack you want to run on it.