The Global Open-Source AI Landscape Is Being Redrawn
A recent observational report from a leading open-source platform sheds new light on the dynamics of global artificial intelligence development. It clearly indicates that the worldwide ecosystem for open-source AI is expanding at a remarkable pace, with contributions from China assuming an increasingly significant role.
The Scale Advantage: China's Parameter Race
The most direct comparison lies in model scale. Data shows that since 2026, for most months, the largest open-source models released by Chinese AI firms have significantly surpassed their U.S. counterparts in parameter count. The upper limit for Chinese models has consistently ranged in the hundreds of billions to trillions of parameters, while U.S. models have mostly remained below a hundred billion during the same period. This lead is not accidental but reflects sustained investment and technical accumulation.
Ecosystem Building: Permissive Licenses Foster Vibrant Communities
Beyond sheer scale, Chinese models are also making strides in building open ecosystems. The report highlights that among models with over 20 billion parameters, more than 80% from China adopt relatively permissive licensing agreements. This openness has a tangible impact: it dramatically lowers the barrier for enterprises and developers to use, modify, and commercially deploy these models, spurring innovation on top of the original foundations.
A prime example is the Qwen model, which has become one of the most pivotal foundational models in the global open-source community. The number of derivative models built upon it has exceeded 150,000, a figure that far surpasses the ecosystem size generated by comparable foundational models from companies like Google and Meta. This demonstrates a strong correlation between technological openness and community vitality.
Shifting Drivers: Hardware Firms Become the New Open-Source Champions
The report uncovers another intriguing shift: within the U.S. market, the primary drivers of open-source model innovation are changing. Historically led by software and internet giants like Google and Meta, the landscape has transformed this year. The top two companies releasing new open-source models are now AMD and NVIDIA.
This shift follows a clear business rationale. For instance, NVIDIA's release of its Nemotron series aims not only to provide a new AI tool but to comprehensively demonstrate the superior synergistic capabilities of its GPU hardware when paired with advanced AI models. It's a "software-sells-hardware" strategy, using open-source excellence to prove hardware value. AMD's involvement focuses more on model conversion and performance optimization, similarly highlighting the advantages of its hardware platforms.
This transition from "software-led" to "hardware-driven" innovation may signal the next phase of AI development: the co-optimization of software and hardware will become critically important, with computational efficiency at the heart of competition.