Hugging Face bet on workflow control

This is a GitHub style bet that control of the workflow will matter more than today’s point product revenue. Hugging Face gives developers the default place to upload model weights, test models, fine tune with standard libraries, and then move into managed inference and enterprise controls. That makes current revenue look like a services and infrastructure wedge into a much larger software position at the center of open AI development.

• The product already behaves like shared infrastructure for open AI. The Hub hosts models, datasets, and demo apps, and the Trainer library has become part of the normal fine tuning workflow for teams building on models like Llama and Mixtral. That kind of habitual usage is what later supports paid seats, governance, and deployment products.
  1 sacra 4 sacra 5 huggingface
• Today’s money comes mostly from managed enterprise usage, not from charging the broad community. The company was at about $70M ARR at the end of 2023, with most revenue tied to managed products sold into large customers, while enterprise features now include SSO, audit logs, private storage, and access controls around deployed endpoints and repositories.
  1 sacra 2 huggingface 6 sacra
• The closest comparables show two different monetization paths. Together AI monetizes usage directly through compute and inference APIs, while Dataiku monetizes a collaborative interface layer used by larger teams inside enterprises. Hugging Face sits between them, with a developer network on top and enterprise deployment and admin products underneath.
  2 huggingface 3 sacra 6 sacra

The next step is moving from being the place where open models are discovered into the place where companies safely run them. If Hugging Face keeps converting community workflow into private deployment, security, and team administration, revenue can expand far beyond hosted inference and start to look like the core system of record for enterprise open source AI.