Open models as soft power and infrastructure
Chinese open-weight models are now serving as a “global technological foundation”, supporting regions including Europe and Southeast Asia. That assessment comes from Zhang Yaqin, founding dean of the Institute for AI Industry Research (AIR) at Tsinghua University, as reported by CGTN. Zhang is one of China’s most prominent AI figures. He served as president of Baidu, where he led major initiatives in autonomous driving, cloud computing and AI, before taking up the AIR role, which sits near the centre of Beijing’s AI policy and industrial ecosystem.
Chinese institutions and companies are leaning into openness as a strategic differentiator. Hundreds of Chinese teams now release model architectures and weights. This creates a shared infrastructure that developers worldwide can inspect, adapt and redeploy. That approach contrasts with the closed model of many US frontier systems. According to recent platform usage data, Chinese open-weight large language models such as Qwen, MiniMax and DeepSeek account for roughly one third of LLM activity on major global routing platforms by mid‑2026.
This open-source tilt also aligns with stated national priorities that call for open ecosystems, deep integration of AI across the real economy, and increased international cooperation. Policy backing and a dense university-industry pipeline have turned institutions like Tsinghua’s AIR into platforms for cross-border collaboration. Chinese models are now widely tested and fine-tuned by start-ups and mid-market firms in Europe, Southeast Asia and other emerging regions.
Chinese officials also frame open-source models as engines of economic growth. China holds the largest share of AI-related patent filings globally; recent studies estimate its share at over half of worldwide AI patent applications. That figure underscores the scale and speed of domestic innovation feeding into these open systems. As a result, China open-source AI is becoming not only a software play, but also a proxy for where future AI value chains may form.
Cross-border data, compute and the next phase of competition
The spread of these models rides on tangible infrastructure. Analysts tracking China’s digital expansion highlight new cross-border data and compute projects. These include multibillion‑dollar data centre investments in Indonesia involving Chinese and regional partners, and emerging data centre and cloud infrastructure projects in several regions, including parts of Central Asia. Both are designed to support training and deployment of open-weight models closer to end users. These investments fit alongside cloud partnerships and submarine cable links under China’s broader digital connectivity push. They give Chinese AI providers local latency, regulatory familiarity and pricing advantages in many emerging markets.
Meanwhile, US policy is moving in the opposite direction. Tighter controls on advanced models and hardware exports, framed around national security, are reinforcing a gated ecosystem for leading US systems. A recent analysis of OpenRouter data finds that by early 2026, Chinese open-weight models accounted for about 61 percent of total token consumption among the platform’s ten most‑used AI models. That signals strong practical adoption even by developers outside China.
For institutional investors, this divergence creates both risk and optionality. Chinese open-source providers can scale quickly in price-sensitive markets. They are embedding their models in government services, finance, logistics and local consumer apps. That positions them as default infrastructure in parts of Southeast Asia, the Middle East, Africa and Central Asia — especially where governments seek digital sovereignty and are wary of over-reliance on any single Western vendor.
However, the same dynamics harden strategic competition. As China open-source AI models become embedded in foreign data flows and decision systems, they shape de facto technical standards — from model interfaces to safety benchmarks. That raises policy and regulatory questions in Europe and elsewhere. It also creates openings for joint ventures, local-language fine-tuning firms and regional data-centre operators aligned with either Chinese or hybrid technology stacks.
The next phase will turn on how far Beijing continues to back openness, how host governments regulate data and security, and whether Western firms respond with more interoperable or open-weight offerings of their own. Investors should watch where Chinese open-weight models such as Qwen, MiniMax and DeepSeek reach roughly one third of LLM usage on major developer platforms by mid‑2026 — those nodes are likely to map the real geography of global AI influence and the most durable returns from China open-source AI.
