Open source AI is booming. Hugging Face explains who profits. | TechCrunch

Open original

This is a short TechCrunch Equity podcast episode where Rebecca Bellan talks with Hugging Face's Irene Solaiman about how companies make money from open AI models and what startups should weigh before building on them. They also cover whether American open model developers can keep pace with Chinese ones.

  • Enterprises, governments and startups are adopting open models for lower costs, more control and more flexibility.
  • The central question is how companies capture value once capable models are freely available.
  • Chinese open models are gaining ground, raising doubts about whether American players can keep up.

Open models are moving from hobbyist territory into the mainstream, and the pitch to enterprises, governments and startups is the same: lower costs, more control, more flexibility. The sticking point is value capture. If capable models are freely downloadable, the question becomes where the money actually comes from. This episode of TechCrunch's Equity podcast takes that question to Irene Solaiman, Chief Policy Officer and SVP of Strategy at Hugging Face, the company that hosts much of the open model ecosystem. The conversation centers on what a startup should weigh before building on an open model. Choosing the right tools and infrastructure comes first, along with the trade-offs that shape the choice between self-hosting, running models on someone else's cloud, or paying for a managed service. Each path shifts responsibility for uptime, security, fine-tuning and compliance, so the cheapest-looking option on paper is not always cheapest in practice. The piece also raises a competitive question: with Chinese open models gaining ground, can American players keep up? That is a live concern for anyone deciding which base model to build on, since the strongest open weights may not come from the country where a company is headquartered. The episode is audio and video, so the specifics of Solaiman's answers, including any figures she cites on adoption or model performance, are only available in the full recording.

Ask about this article