During the early weeks of summer, the AI community was glued to Anthropic’s newest frontier models and Washington’s attempts to regulate who could access them. Yet behind the headlines, developers kept building open‑weight models without waiting for permission from the likes of Anthropic or OpenAI.

Open Models Capture a Growing Share of the Market

Data from Vercel shows that open‑weight models now absorb the bulk of high‑volume AI infrastructure, while closed models sit in a higher‑cost, premium tier. In June, open models handled nearly one‑third of all AI requests on the platform. Although this metric excludes sessions hosted by major labs—likely the majority of OpenAI and Anthropic traffic—the expanding share of open‑source models raises a critical question: how much relevance do frontier models retain when most production AI runs on cheaper, customizable alternatives?

Enterprise Shift: Ownership Over Rental

Hugging Face CEO Clem Delangue told the recent episode of “Equity” that companies increasingly prefer to own their AI models rather than rent black‑box APIs. The steep scaling costs of closed frontier models have made this shift inevitable. Delangue notes that a new repository is created on Hugging Face every seven seconds, now hosting almost three million public models and one million public datasets. This paints a picture far from a “one model to rule them all” scenario; instead, enterprises are deploying a portfolio of models, many of which are fine‑tuned for specific use‑cases.

China’s Accelerating Open‑Weight Releases

Chinese AI labs are consistently releasing powerful open‑weight models that are cheaper to deploy and easier to customize than their closed counterparts. The latest example is Beijing‑based Z.ai’s GLM‑5.2, which excels at agentic coding and competes directly with Anthropic’s latest offerings on security‑vulnerability detection. These releases undercut the economics of proprietary AI that U.S. firms have poured billions into, signaling a shift in global AI power dynamics.

Transparency vs. Control: The Ongoing Debate

Microsoft CEO Satya Nadella recently warned against vendor lock‑in, emphasizing data control as a primary concern for enterprises. Conversely, Anthropic CEO Dario Amodei cautioned that scaling open model weights could become dangerous because they become hard to regulate once released. Delanguu argues the biggest AI risk is concentration of power, and the safest path forward is to level the playing field through transparency. Restricting powerful models behind closed doors does not eliminate risk; it merely concentrates technology in the hands of a few, creating asymmetry of power and capabilities.