Chinese AI firms have taken the lead in the open‑source arena, accounting for 41% of all model downloads on Hugging Face this spring—surpassing U.S. models for the first time. The surge isn’t limited to one platform; on OpenRouter, the six most popular models are all open‑weight offerings from Chinese companies such as Tencent, Xiaomi, DeepSeek, MiniMax and Z.ai, relegating Anthropic’s Claude Opus 4.7 to seventh place.

Developers are not waiting for permission from the likes of OpenAI or Anthropic. They are building, deploying, and scaling their own solutions with models that cost less to run and can be tailored to specific tasks. Data from Vercel shows open‑weight models processed almost a third of all AI requests in June, positioning them as the volume‑heavy, low‑cost layer beneath premium closed‑source services.

Hugging Face, the platform that hosts nearly three million public models, sees this trend reflected in its own activity. A new repository appears every seven seconds, and half of Fortune 500 companies now use the service to launch private or open‑source models. CEO Clem Delangue told the Equity podcast that enterprises increasingly prefer owning their AI capabilities rather than renting a black‑box API.

"If you’re an AI company or a technology company, you don’t want to outsource your core capabilities to another company, to a black box API that you don’t control," Delangue said. "You need visibility, ownership and the ability to customize." The sentiment echoes Microsoft CEO Satya Nadella’s warning against single‑provider lock‑in, emphasizing that data control should be a primary concern for businesses adopting AI.

Chinese labs are fueling the momentum with a steady stream of powerful releases. Z.ai’s recent GLM‑5.2 model, for example, rivals Anthropic’s latest offerings in identifying security vulnerabilities while remaining cheaper to deploy. Each new open‑weight model undercuts the economics of proprietary AI that U.S. firms have poured billions into.

The rise of open‑source models also reignites debate over safety and accessibility. Anthropic’s Dario Amodei cautions that widely releasing powerful weights could be dangerous, arguing that once models are out, they become hard to control. Critics fear that open models may be repurposed by bad actors for disinformation or cyber‑warfare.

Delangue pushes back, suggesting that concentration of power poses a greater risk. "The biggest risk in AI is concentration of power," he said. "Leveling the playing field and creating transparency makes the world safer." He points out that open models allow defenders to patch known vulnerabilities more quickly than closed systems, where security flaws remain hidden.

While the frontier of AI research still belongs to large labs, the bulk of production workloads appears to be shifting toward cheaper, customizable alternatives. As enterprises prioritize control, cost and flexibility, open‑source models are poised to become the backbone of everyday AI applications, leaving closed‑source offerings to niche, high‑value tasks.

Cet article a été rédigé avec l'assistance de l'IA.
News Factory APP - actualités agentiques pour booster votre SEO et AEO.