China’s Iluvatar CoreX’s first AI computing supernode squeezes most GPUs in 1 system

The Tiangai 300 IXL144 system is designed for large-scale AI model training and inference, packing 144 GPUs and 18 CPUs into a single cabinet. This dense integration, alongside 10.1 terabytes of memory, suggests an effort to maximize computing resources within a compact physical footprint, which could be beneficial for organizations managing substantial AI workloads.

Iluvatar CoreX's claim of a 40% reduction in cost per 1 million valid tokens compared to "mainstream international alternatives" highlights a potential economic advantage. If validated, such a cost efficiency could influence hardware procurement decisions, particularly for entities seeking to optimize spending on AI infrastructure.

The launch occurs amidst US export controls affecting access to advanced chips. By developing a deeply integrated supernode system domestically, Iluvatar CoreX is positioned to offer a local option for scaling computing power. This provides an alternative for users in China who may face restrictions on international chip supplies, addressing a specific market demand driven by geopolitical factors.