The artificial intelligence (AI) race is witnessing a major shift as enterprises gradually turn away from expensive closed APIs to adopt open-weight models. According to the latest data from Hugging Face and application development platforms, open models, especially from China, are dominating actual production traffic in enterprise environments. This shift raises a major question about the dominant role of proprietary frontier models that cost billions of dollars to develop.
Diễn biến chi tiết
According to Hugging Face's spring 2026 report, open models from Chinese companies accounted for 41% of downloads on the platform, officially surpassing U.S. models. On OpenRouter, the six most popular models are currently all open-weight from major Chinese firms such as Tencent, Xiaomi, DeepSeek, MiniMax, and Z.ai, pushing Anthropic's Claude 4.7 Opus down to seventh place. Meanwhile, data from Vercel's AI Gateway shows that open models handled nearly a third of all AI requests in June 2026. This trend reflects enterprises seeking cheaper, highly customizable alternatives rather than relying entirely on third-party providers.
Phân tích kỹ thuật & Công nghệ
The rise of open models is driven by the ability to optimize infrastructure costs and gain technical control. Instead of using "black box" APIs that cannot be deeply customized, system engineers prefer open-weight models because they can directly fine-tune parameters and embed internal data via RAG techniques without fear of data leaks. For instance, Z.ai's newly released GLM-5.2 open model excels at agentic coding and competes directly with top closed models in identifying security vulnerabilities. Running models on private infrastructure helps companies optimize token consumption and avoid runaway scaling costs.
Ý kiến chuyên gia & Nhận định
Hugging Face CEO Clem Delangue predicted that in a few years, large frontier models will only be used for experimentation or high-value tasks, while most production workloads will run on private or open-source models. He emphasized: "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." Even Microsoft CEO Satya Nadella warned against vendor lock-in, criticizing closed model providers for imposing restrictive terms on distillation and learning from customer data, urging the distribution of learning infrastructure so firms can control their own data loops.
Tác động & Tương lai
The boom of open models also deepens the debate over safety. Contrary to concerns from closed-model advocates like Anthropic CEO Dario Amodei about technology falling into the wrong hands, Clem Delangue argued that the biggest risk in AI is the concentration of power. Keeping models behind closed doors does not increase safety but simply creates asymmetry of capabilities. For the tech community in Vietnam, the availability of high-quality open models opens up great opportunities to access and build specialized AI solutions at optimized costs, free from the technical and policy barriers of Western tech giants.