China’s innovation thrives on openness, even under authoritarian rule |
Understanding the reality of “innovation” in China and properly evaluating its technological capabilities is extremely difficult.
One reason lies in the ambiguity of the word “freedom” in Chinese society. While freedom of speech, especially speech concerning the legitimacy of the regime, is strictly suppressed, the industrial sector is marked by vigorous competition and technological innovation among private companies.
Accurately assessing this coexistence is no easy task. This briefing will therefore focus on intense corporate competition premised on the “free use” of technology — that is, technological openness — and will examine how it has affected China’s innovation and industrial competitiveness.
When people hear “technological openness,” many likely think of open-source software such as Linux. In fact, open-source software can also help address economic security challenges.
In recent years, both China’s public and private sectors have focused on RISC-V, an open-source instruction set architecture. RISC-V has likely played a role in facilitating domestic semiconductor development in China, since U.S.-led restrictions on exports of advanced semiconductors and related technologies to Chinese companies have also affected the processor architecture industry, where the U.K.-based Arm, a leading licensor of processor designs, holds a major share.
The power of technological openness has been even more evident in the rapid progress of AI developed by Chinese companies. After OpenAI released ChatGPT to the public in November 2022 and global attention on generative AI surged, the Chinese government quickly advocated a “battle of a hundred models.” By mid-2023, Chinese engineers had developed more than 130 large language models, accounting for roughly 40% of the global total. Most were open-weight models.
In January 2025, the success of DeepSeek’s open-weight LLMs, V3 and R1, shocked the world. In July 2026, Moonshot AI developed Kimi K3, a large language model that outperformed Anthropic’s top model, Claude Fable 5, in some benchmarks such........