China Launches National AI for Scientific Research
China has launched a new national AI system connected to its National Supercomputing Network. The system is reportedly designed to conduct independent scientific research, signaling a strategic state-level priority to anchor AI development in foundational infrastructure. This move underscores the country's intent to build sovereign capabilities in advanced AI research and applications.
- The system, part of China's "AI for Science" initiative, integrates over 30 supercomputing centers via the National Supercomputing Network (SCNet) and supports nearly 100 scientific workflows in fields like materials science and biotechnology. This initiative was preceded by a March 2023 plan from the Ministry of Science and Technology to drive AI adoption in basic sciences. - This national strategy is mirrored in the private sector, where Chinese tech giants have developed advanced multi-agent frameworks. Tencent's Hunyuan, for instance, acts as the AI backbone for WeChat's enterprise ecosystem and its Agent Runtime is a full agentic AI deployment fabric. Similarly, Alibaba's Qwen AI has over 80 variants, including models specifically for reasoning and agent-tool use. - The launch is a direct response to international competition, particularly the United States' "Genesis Mission" initiative, which also aims to harness federal supercomputers and government research data to train powerful AI agents for scientific discovery. - Beijing is aggressively fostering a supportive local ecosystem for such advancements, with a plan to expand the city's core AI industry to over 1 trillion yuan (US$142.5 billion) within two years. The plan includes developing a domestic AI computing cluster with more than 100,000 chips and cultivating over 20 AI unicorn companies. - The regulatory environment for deploying AI agents in China is also evolving. While a single comprehensive AI law was removed from the 2025 legislative schedule, the government is pursuing a phased approach with targeted measures, pilots, and technical standards for model evaluation, data governance, and cybersecurity testing. - Chinese startups are already launching fully autonomous agents with multi-agent architectures that break down large tasks for sub-agents to execute. For example, Manus AI, created by the startup Butterfly Effect, is designed to independently manage complex workflows, from analyzing documents to building and launching a website. - Recent large language model releases in China are increasingly focused on enabling more sophisticated agentic workflows. Alibaba's Qwen3.5, Zhipu AI's GLM-5, and ByteDance's Doubao are all positioned as foundational models for digital agents capable of multi-step reasoning, planning, and complex tool use.
