US vs China in AI: How Competing Rules, Chips, and Ecosystems Are Shaping a New Digital Cold War (Part Two)
Today, it is evident that China has successfully built a highly competitive AI ecosystem. The country’s open-source AI framework is advancing rapidly and purposefully. In recent months, Chinese laboratories have dominated the rankings on open model evaluation platforms such as LM Arena.
Leading the charge is Kimi-K2 from Moonshot AI, which has overtaken DeepSeek as the most-voted open model on the platform. It’s clear that while Meta lost the open source battle in the US, Chinese labs compete head-to-head with their Western counterparts. It’s also being widely adopted across Chinese developer platforms, with user feedback loops and community fine-tuning driving rapid iteration.
Qwen3 from Alibaba is gaining traction after the team made a deliberate move to split its architecture into separate Instruct and Thinking variants. This shift came after ongoing criticism from the community regarding earlier hybrid models that muddled the distinction between following instructions and maintaining reasoning fidelity.
At the heart of these technological advancements lies a robust collaboration between the public and private sectors. Chinese ministries and regional governments are providing computing subsidies, promoting standardised benchmarks, and subtly enforcing alignment standards related to ideological conformity. With each new release, the Chinese ecosystem demonstrates its ability to accelerate progress, expand reach, and develop models that address both domestic needs and international strategic objectives. Applications are not the sole area of China’s emerging supremacy. Currently, China either matches or surpasses the US in AI publication output and has effectively eliminated the historical lag in addressing new AI topics.
A 2023 study, Has China caught up to the US in AI research?, finds China outproduces the US in AI-related papers but still trails on quality metrics like citations; nevertheless, the gap has narrowed dramatically, and China may now be on an “independent trajectory” rather than simply imitating.
A 2024 analysis in Scientific Reports shows that China and the US produce the world’s most impactful AI research overall, and that collaboration between Chinese and US scientists tends to generate more highly cited work than either country alone.
China now leads the world in AI patent volume. In 2024, it had approximately 188,000 AI patent filings, compared with roughly 8,600 granted AI patents in the US, and some analyses suggest that China accounts for over 70% of global AI patent filings in recent years.
At the chip and infrastructure level, China remains at a disadvantage. A 2025 MERICS report highlights that although China has created its own AI chips and is advocating for “independent and controllable” AI across all layers, its AI accelerators still fall short of the performance offered by Nvidia’s top GPUs. Furthermore, US export restrictions on advanced semiconductors and equipment continue to pose a significant limitation.
Higher in the stack—models and applications—China is much closer. The same report concludes that in large language models and applied AI, Chinese firms like Baidu, Alibaba, Tencent and newer players such as DeepSeek are now “closing in on the US,” helped by a protected home market, huge data volume, and deep engagement with global open-source communities.
The trajectory is clear: through massive state-backed investment, talent programs, and a high-volume patenting strategy, China has moved from laggard to co-leader in AI R&D and IP within about a decade, turning the US-China AI race into a genuine two-pole system rather than a US-only frontier.