Efforts to Bypass Chip Restrictions

Reports indicate that Moonshot, a Chinese artificial intelligence laboratory, has explored utilizing data centers located in Europe to gain access to Nvidia chips. This move comes as U.S. export controls continue to restrict the flow of advanced semiconductor technology to Chinese firms.

The reported interest in European infrastructure highlights the challenges faced by Chinese AI developers attempting to secure the high-end hardware necessary for training large-scale models. While the reports regarding Moonshot’s specific search for European capacity remain unverified by independent confirmation, they reflect a broader trend of Chinese firms seeking alternative routes to maintain computational power amidst tightening trade restrictions.

Domestic Hardware and Infrastructure Developments

Despite international sanctions, Chinese technology firms are increasingly investing in domestic hardware and energy-efficient infrastructure. Baidu has reported the successful operation of a cluster comprising 30,000 of its self-developed P800 Kunlun chips, stating the cluster is capable of training models similar to DeepSeek.

In terms of power management, China has deployed what it describes as the world’s first 800-volt DC power system for a massive AI data center. The system is designed to increase conversion efficiency and reduce the physical footprint required for high-density computing. Additionally, China has launched its first AI data center powered entirely by green electricity.

Competitive Benchmarks and Model Performance

The competitive landscape between Chinese and U.S. AI models remains active, with several firms reporting significant performance metrics:

  • Moonshot AI: The company’s Kimi K2 model has reportedly outperformed GPT-4 in specific key benchmarks.
  • MiniMax: The firm stated that its M1 AI model costs approximately 200 times less to train than OpenAI’s GPT-4.
  • Kuaishou Technology: The company has unveiled an AI video generator intended to compete with OpenAI’s Sora.
  • Alibaba: The firm has released the open-source Qwen3 model, including Qwen3-Coder, which is designed for software development and complex coding workflows.
  • DeepSeek: Recent reports suggest the DeepSeek-V3.1 model has been optimized for Chinese-made chips and offers improved processing speeds over its predecessor.

Expanding Ecosystem and Research

Chinese firms are increasingly forming alliances to build a domestic AI ecosystem to mitigate the impact of U.S. curbs. This includes the development of specialized tools, such as the ScienceOne platform from the Chinese Academy of Sciences (CAS), which features the S1-ToolChain to coordinate over 300 scientific tools.

In the specialized field of robotics, Alibaba’s RynnBrain has reportedly set 16 robotics records, outperforming certain models from Google and Nvidia. Meanwhile, ByteDance has unveiled Seedream 4.0, an AI image model. While some claims regarding the superiority of specific models—such as GigaAI’s GigaWorld-1—circulate on social media platforms like Reddit, these performance claims lack independent, third-party verification at this time.