China's AI infrastructure development is undergoing a profound transformation. Recently, the State Council issued a notice implementing a "domestic product standard" in government procurement, granting a 20% price evaluation preference for locally manufactured equipment. Industry experts believe this will prioritize domestic hardware in critical infrastructure projects like data centers. Concurrently, reports suggest China has banned foreign AI chips in newly built state-owned data centers, with projects under 30% completion required to remove installed foreign chips or cancel orders.

Current data indicates non-domestic chips now account for a historic low in China's AI infrastructure. For instance, U.S. export restrictions have drastically reduced NVIDIA's market share in China, with some executives admitting its computing power presence has "nearly vanished." The "domestic-first" policy is accelerating localization in AI infrastructure, enhancing supply chain stability.

Amid this shift, Sugon's scaleX640—a domestically developed super-node—has unveiled a 20x leap in computing density. Designed for trillion-parameter AI models, the scaleX640 features 630 PFlops performance (double mainstream alternatives) and packs 640 accelerator cards per rack (20x higher density than typical 32-card setups). It also boasts advancements in HBM bandwidth and energy efficiency.

Notably, Sugon’s cooling and power technologies reportedly match NVIDIA’s projected 2027 NVL576 architecture, achieving a 9x integration boost over current NVL72 systems. With full compatibility for mainstream AI ecosystems (including CUDA alternatives), domestic solutions like scaleX640 are bridging the performance gap amid U.S. export controls.

From policy tailwinds to technological breakthroughs, China’s AI infrastructure and homegrown super-nodes are converging to solidify the foundation for its artificial intelligence ambitions.