Hugging Face CEO says open source AI is winning on cost and control

Clem Delangue said companies are reaching a point where renting frontier AI from a small group of large vendors no longer makes economic sense, and that open source models are winning on cost and control. Hugging Face’s Spring 2026 State of Open Source report said more than 30% of the Fortune 500 now maintain verified accounts on the platform, a signal that enterprise buyers are moving beyond early experiments and into broader deployment.

Hugging Face, often described as a GitHub for AI, said more than 50,000 organizations are now using the platform. The company said its ecosystem reached 13 million users, more than 2 million models and more than 500,000 datasets in 2025. Those figures point to a platform that has become part of enterprise AI infrastructure, not just a place to browse models.

The shift Delangue described is also showing up in the market. Reuters and CNBC reported in late June and early July 2026 that AI costs were rising and pushing some companies toward cheaper model alternatives. CNBC said on July 7, 2026, that Chinese-built models including DeepSeek and Z.ai were gaining traction among U.S. companies because they were significantly cheaper to use. That price pressure has made the question less about ideology and more about procurement, margins and how much control a company wants over the systems it deploys.

OpenAI has responded to the same pressure on the closed-model side. Its enterprise materials say enterprise AI is entering a phase in which large organizations are translating AI capabilities into scaled use cases, and the company has added enterprise-grade features and cost-management controls for API customers. That makes the competitive field clearer: open source vendors are pitching lower long-term costs, customization and ownership, while closed-model providers are leaning harder on governance, spending controls and enterprise administration.

Delangue’s warning is that concentration itself carries risk if a handful of firms come to control the most important AI systems. The companies buying these tools are increasingly weighing not just model quality, but who can inspect the system, modify it and keep the data and deployment inside their own stack.