The Strange Business of Hugging Face

Why are there so many models that no one uses?

Welcome to the latest edition of Buy the Rumor; Sell the News. We’re closing in on 2,000 subscribers, including institutional allocators, venture capitalists, litigators, senior executives and entrepreneurs. Thank you to all who have subscribed!

In today’s post, I take a look at Hugging Face. Thousands of large language models hosted, the vast majority of which seem to have no users. What’s the point? How does the business make money? And what’s its moat?

If you want to connect with me directly, my contact information is at the end of this post.

---

Hugging Face is, on its face, a deeply puzzling business. It’s a sprawling, messy bazaar of thousands of AI models, most of which are little more than half-baked science experiments that no serious enterprise would deploy. So why has this company raised hundreds of millions, become the darling of the open-weights crowd, and found itself mentioned in nearly every conversation about trustworthy AI?

The short answer is that Hugging Face isn’t really about the models. It’s about owning the coordination layer for how open machine learning happens. The models are incidental; what matters is that Hugging Face has become the default namespace, workflow, and social graph for everything from toy GPT fine-tunes to billion-dollar enterprise deployments.